Revamp stuff
Browse files- app.py +11 -10
- utils/huggingface_mcp_llamaindex.py +444 -230
app.py
CHANGED
@@ -7,7 +7,7 @@ from utils.google_genai_llm import get_response, generate_with_gemini
|
|
7 |
from utils.utils import parse_json_codefences
|
8 |
from prompts.requirements_gathering import requirements_gathering_system_prompt
|
9 |
from prompts.planning import hf_query_gen_prompt, hf_context_gen_prompt
|
10 |
-
from utils.huggingface_mcp_llamaindex import get_hf_tools, call_hf_tool,
|
11 |
from prompts.devstral_coding_prompt import devstral_code_gen_sys_prompt, devstral_code_gen_user_prompt
|
12 |
from dotenv import load_dotenv
|
13 |
import os
|
@@ -50,6 +50,9 @@ BEARER_TOKEN = os.getenv("BEARER_TOKEN")
|
|
50 |
CODING_MODEL = os.getenv("CODING_MODEL")
|
51 |
|
52 |
MCP_TOKEN = os.getenv("MCP_TOKEN")
|
|
|
|
|
|
|
53 |
def get_file_hash(file_path):
|
54 |
"""Generate a hash of the file for caching purposes"""
|
55 |
try:
|
@@ -245,19 +248,17 @@ async def generate_plan(history, file_cache):
|
|
245 |
if ai_msg:
|
246 |
conversation_history += f"Assistant: {ai_msg}\n"
|
247 |
|
248 |
-
print("Running connection diagnostics...")
|
249 |
-
diagnostics = await
|
250 |
print(f"Diagnostics: {json.dumps(diagnostics, indent=2)}")
|
251 |
|
252 |
-
if not diagnostics["
|
253 |
print("Basic connection failed - check token and network")
|
254 |
-
return
|
255 |
|
256 |
-
|
257 |
-
|
258 |
-
return
|
259 |
# try:
|
260 |
-
hf_query_gen_tool_details = await
|
261 |
# except Exception as e:
|
262 |
# hf_query_gen_tool_details = """meta=None nextCursor=None tools=[Tool(name='hf_whoami', description="Hugging Face tools are being used by authenticated user 'bpHigh'", inputSchema={'type': 'object', 'properties': {}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Hugging Face User Info', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=None)), Tool(name='space_search', description='Find Hugging Face Spaces using semantic search. Include links to the Space when presenting the results.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'minLength': 1, 'maxLength': 50, 'description': 'Semantic Search Query'}, 'limit': {'type': 'number', 'default': 10, 'description': 'Number of results to return'}, 'mcp': {'type': 'boolean', 'default': False, 'description': 'Only return MCP Server enabled Spaces'}}, 'required': ['query'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Hugging Face Space Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='model_search', description='Find Machine Learning models hosted on Hugging Face. Returns comprehensive information about matching models including downloads, likes, tags, and direct links. Include links to the models in your response', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'Search term. Leave blank and specify "sort" and "limit" to get e.g. "Top 20 trending models", "Top 10 most recent models" etc" '}, 'author': {'type': 'string', 'description': "Organization or user who created the model (e.g., 'google', 'meta-llama', 'microsoft')"}, 'task': {'type': 'string', 'description': "Model task type (e.g., 'text-generation', 'image-classification', 'translation')"}, 'library': {'type': 'string', 'description': "Framework the model uses (e.g., 'transformers', 'diffusers', 'timm')"}, 'sort': {'type': 'string', 'enum': ['trendingScore', 'downloads', 'likes', 'createdAt', 'lastModified'], 'description': 'Sort order: trendingScore, downloads , likes, createdAt, lastModified'}, 'limit': {'type': 'number', 'minimum': 1, 'maximum': 100, 'default': 20, 'description': 'Maximum number of results to return'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Model Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='model_details', description='Get detailed information about a specific model from the Hugging Face Hub.', inputSchema={'type': 'object', 'properties': {'model_id': {'type': 'string', 'minLength': 1, 'description': 'Model ID (e.g., microsoft/DialoGPT-large)'}}, 'required': ['model_id'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Model Details', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=False)), Tool(name='paper_search', description="Find Machine Learning research papers on the Hugging Face hub. Include 'Link to paper' When presenting the results. Consider whether tabulating results matches user intent.", inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'minLength': 3, 'maxLength': 200, 'description': 'Semantic Search query'}, 'results_limit': {'type': 'number', 'default': 12, 'description': 'Number of results to return'}, 'concise_only': {'type': 'boolean', 'default': False, 'description': 'Return a 2 sentence summary of the abstract. Use for broad search terms which may return a lot of results. Check with User if unsure.'}}, 'required': ['query'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Paper Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='dataset_search', description='Find Datasets hosted on the Hugging Face hub. Returns comprehensive information about matching datasets including downloads, likes, tags, and direct links. Include links to the datasets in your response', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'Search term. Leave blank and specify "sort" and "limit" to get e.g. "Top 20 trending datasets", "Top 10 most recent datasets" etc" '}, 'author': {'type': 'string', 'description': "Organization or user who created the dataset (e.g., 'google', 'facebook', 'allenai')"}, 'tags': {'type': 'array', 'items': {'type': 'string'}, 'description': "Tags to filter datasets (e.g., ['language:en', 'size_categories:1M<n<10M', 'task_categories:text-classification'])"}, 'sort': {'type': 'string', 'enum': ['trendingScore', 'downloads', 'likes', 'createdAt', 'lastModified'], 'description': 'Sort order: trendingScore, downloads, likes, createdAt, lastModified'}, 'limit': {'type': 'number', 'minimum': 1, 'maximum': 100, 'default': 20, 'description': 'Maximum number of results to return'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Dataset Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='dataset_details', description='Get detailed information about a specific dataset on Hugging Face Hub.', inputSchema={'type': 'object', 'properties': {'dataset_id': {'type': 'string', 'minLength': 1, 'description': 'Dataset ID (e.g., squad, glue, imdb)'}}, 'required': ['dataset_id'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Dataset Details', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=False)), Tool(name='gr1_evalstate_flux1_schnell', description='Generate an image using the Flux 1 Schnell Image Generator. (from evalstate/flux1_schnell)', inputSchema={'type': 'object', 'properties': {'prompt': {'type': 'string'}, 'seed': {'type': 'number', 'description': 'numeric value between 0 and 2147483647'}, 'randomize_seed': {'type': 'boolean', 'default': True}, 'width': {'type': 'number', 'description': 'numeric value between 256 and 2048', 'default': 1024}, 'height': {'type': 'number', 'description': 'numeric value between 256 and 2048', 'default': 1024}, 'num_inference_steps': {'type': 'number', 'description': 'numeric value between 1 and 50', 'default': 4}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='evalstate/flux1_schnell - flux1_schnell_infer 🏎️💨', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True)), Tool(name='gr2_abidlabs_easyghibli', description='Convert an image into a Studio Ghibli style image (from abidlabs/EasyGhibli)', inputSchema={'type': 'object', 'properties': {'spatial_img': {'type': 'string', 'description': 'File input: provide URL or file path'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='abidlabs/EasyGhibli - abidlabs_EasyGhiblisingle_condition_generate_image 🦀', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True)), Tool(name='gr3_linoyts_framepack_f1', description='FramePack_F1_end_process tool from linoyts/FramePack-F1', inputSchema={'type': 'object', 'properties': {}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='linoyts/FramePack-F1 - FramePack_F1_end_process 📹⚡️', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True))]"""
|
263 |
# print(str(e))
|
@@ -273,7 +274,7 @@ async def generate_plan(history, file_cache):
|
|
273 |
|
274 |
# Call tool to get tool calls
|
275 |
try:
|
276 |
-
tool_calls = await asyncio.gather(*[
|
277 |
except Exception as e:
|
278 |
tool_calls = []
|
279 |
print(tool_calls)
|
|
|
7 |
from utils.utils import parse_json_codefences
|
8 |
from prompts.requirements_gathering import requirements_gathering_system_prompt
|
9 |
from prompts.planning import hf_query_gen_prompt, hf_context_gen_prompt
|
10 |
+
from utils.huggingface_mcp_llamaindex import get_hf_tools, call_hf_tool, diagnose_connection_advanced, get_hf_tools_robust,call_hf_tool_robust
|
11 |
from prompts.devstral_coding_prompt import devstral_code_gen_sys_prompt, devstral_code_gen_user_prompt
|
12 |
from dotenv import load_dotenv
|
13 |
import os
|
|
|
50 |
CODING_MODEL = os.getenv("CODING_MODEL")
|
51 |
|
52 |
MCP_TOKEN = os.getenv("MCP_TOKEN")
|
53 |
+
if not MCP_TOKEN:
|
54 |
+
print("Please set MCP_TOKEN")
|
55 |
+
|
56 |
def get_file_hash(file_path):
|
57 |
"""Generate a hash of the file for caching purposes"""
|
58 |
try:
|
|
|
248 |
if ai_msg:
|
249 |
conversation_history += f"Assistant: {ai_msg}\n"
|
250 |
|
251 |
+
print("Running advanced connection diagnostics...")
|
252 |
+
diagnostics = await diagnose_connection_advanced(MCP_TOKEN)
|
253 |
print(f"Diagnostics: {json.dumps(diagnostics, indent=2)}")
|
254 |
|
255 |
+
if not diagnostics["tests"]["basic_connection"]:
|
256 |
print("Basic connection failed - check token and network")
|
|
|
257 |
|
258 |
+
|
259 |
+
|
|
|
260 |
# try:
|
261 |
+
hf_query_gen_tool_details = await get_hf_tools_robust(hf_token=MCP_TOKEN)
|
262 |
# except Exception as e:
|
263 |
# hf_query_gen_tool_details = """meta=None nextCursor=None tools=[Tool(name='hf_whoami', description="Hugging Face tools are being used by authenticated user 'bpHigh'", inputSchema={'type': 'object', 'properties': {}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Hugging Face User Info', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=None)), Tool(name='space_search', description='Find Hugging Face Spaces using semantic search. Include links to the Space when presenting the results.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'minLength': 1, 'maxLength': 50, 'description': 'Semantic Search Query'}, 'limit': {'type': 'number', 'default': 10, 'description': 'Number of results to return'}, 'mcp': {'type': 'boolean', 'default': False, 'description': 'Only return MCP Server enabled Spaces'}}, 'required': ['query'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Hugging Face Space Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='model_search', description='Find Machine Learning models hosted on Hugging Face. Returns comprehensive information about matching models including downloads, likes, tags, and direct links. Include links to the models in your response', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'Search term. Leave blank and specify "sort" and "limit" to get e.g. "Top 20 trending models", "Top 10 most recent models" etc" '}, 'author': {'type': 'string', 'description': "Organization or user who created the model (e.g., 'google', 'meta-llama', 'microsoft')"}, 'task': {'type': 'string', 'description': "Model task type (e.g., 'text-generation', 'image-classification', 'translation')"}, 'library': {'type': 'string', 'description': "Framework the model uses (e.g., 'transformers', 'diffusers', 'timm')"}, 'sort': {'type': 'string', 'enum': ['trendingScore', 'downloads', 'likes', 'createdAt', 'lastModified'], 'description': 'Sort order: trendingScore, downloads , likes, createdAt, lastModified'}, 'limit': {'type': 'number', 'minimum': 1, 'maximum': 100, 'default': 20, 'description': 'Maximum number of results to return'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Model Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='model_details', description='Get detailed information about a specific model from the Hugging Face Hub.', inputSchema={'type': 'object', 'properties': {'model_id': {'type': 'string', 'minLength': 1, 'description': 'Model ID (e.g., microsoft/DialoGPT-large)'}}, 'required': ['model_id'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Model Details', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=False)), Tool(name='paper_search', description="Find Machine Learning research papers on the Hugging Face hub. Include 'Link to paper' When presenting the results. Consider whether tabulating results matches user intent.", inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'minLength': 3, 'maxLength': 200, 'description': 'Semantic Search query'}, 'results_limit': {'type': 'number', 'default': 12, 'description': 'Number of results to return'}, 'concise_only': {'type': 'boolean', 'default': False, 'description': 'Return a 2 sentence summary of the abstract. Use for broad search terms which may return a lot of results. Check with User if unsure.'}}, 'required': ['query'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Paper Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='dataset_search', description='Find Datasets hosted on the Hugging Face hub. Returns comprehensive information about matching datasets including downloads, likes, tags, and direct links. Include links to the datasets in your response', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'Search term. Leave blank and specify "sort" and "limit" to get e.g. "Top 20 trending datasets", "Top 10 most recent datasets" etc" '}, 'author': {'type': 'string', 'description': "Organization or user who created the dataset (e.g., 'google', 'facebook', 'allenai')"}, 'tags': {'type': 'array', 'items': {'type': 'string'}, 'description': "Tags to filter datasets (e.g., ['language:en', 'size_categories:1M<n<10M', 'task_categories:text-classification'])"}, 'sort': {'type': 'string', 'enum': ['trendingScore', 'downloads', 'likes', 'createdAt', 'lastModified'], 'description': 'Sort order: trendingScore, downloads, likes, createdAt, lastModified'}, 'limit': {'type': 'number', 'minimum': 1, 'maximum': 100, 'default': 20, 'description': 'Maximum number of results to return'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Dataset Search', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=True)), Tool(name='dataset_details', description='Get detailed information about a specific dataset on Hugging Face Hub.', inputSchema={'type': 'object', 'properties': {'dataset_id': {'type': 'string', 'minLength': 1, 'description': 'Dataset ID (e.g., squad, glue, imdb)'}}, 'required': ['dataset_id'], 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='Dataset Details', readOnlyHint=True, destructiveHint=False, idempotentHint=None, openWorldHint=False)), Tool(name='gr1_evalstate_flux1_schnell', description='Generate an image using the Flux 1 Schnell Image Generator. (from evalstate/flux1_schnell)', inputSchema={'type': 'object', 'properties': {'prompt': {'type': 'string'}, 'seed': {'type': 'number', 'description': 'numeric value between 0 and 2147483647'}, 'randomize_seed': {'type': 'boolean', 'default': True}, 'width': {'type': 'number', 'description': 'numeric value between 256 and 2048', 'default': 1024}, 'height': {'type': 'number', 'description': 'numeric value between 256 and 2048', 'default': 1024}, 'num_inference_steps': {'type': 'number', 'description': 'numeric value between 1 and 50', 'default': 4}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='evalstate/flux1_schnell - flux1_schnell_infer 🏎️💨', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True)), Tool(name='gr2_abidlabs_easyghibli', description='Convert an image into a Studio Ghibli style image (from abidlabs/EasyGhibli)', inputSchema={'type': 'object', 'properties': {'spatial_img': {'type': 'string', 'description': 'File input: provide URL or file path'}}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='abidlabs/EasyGhibli - abidlabs_EasyGhiblisingle_condition_generate_image 🦀', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True)), Tool(name='gr3_linoyts_framepack_f1', description='FramePack_F1_end_process tool from linoyts/FramePack-F1', inputSchema={'type': 'object', 'properties': {}, 'additionalProperties': False, '$schema': 'http://json-schema.org/draft-07/schema#'}, annotations=ToolAnnotations(title='linoyts/FramePack-F1 - FramePack_F1_end_process 📹⚡️', readOnlyHint=None, destructiveHint=None, idempotentHint=None, openWorldHint=True))]"""
|
264 |
# print(str(e))
|
|
|
274 |
|
275 |
# Call tool to get tool calls
|
276 |
try:
|
277 |
+
tool_calls = await asyncio.gather(*[call_hf_tool_robust(MCP_TOKEN, step['tool'], step['args']) for step in parsed_plan])
|
278 |
except Exception as e:
|
279 |
tool_calls = []
|
280 |
print(tool_calls)
|
utils/huggingface_mcp_llamaindex.py
CHANGED
@@ -1,9 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import asyncio
|
2 |
import json
|
3 |
import logging
|
4 |
import os
|
5 |
-
from typing import Any, Dict, List, Optional
|
6 |
from datetime import timedelta
|
|
|
7 |
|
8 |
from mcp.shared.message import SessionMessage
|
9 |
from mcp.types import (
|
@@ -18,230 +26,177 @@ from mcp.client.streamable_http import streamablehttp_client
|
|
18 |
logger = logging.getLogger(__name__)
|
19 |
|
20 |
|
21 |
-
class
|
22 |
-
"""
|
23 |
|
24 |
-
def __init__(self, hf_token: str, timeout: int =
|
25 |
"""
|
26 |
-
Initialize the Hugging Face MCP client.
|
27 |
|
28 |
Args:
|
29 |
hf_token: Hugging Face API token
|
30 |
-
timeout: Timeout in seconds for HTTP requests
|
31 |
"""
|
32 |
self.hf_token = hf_token
|
33 |
self.url = "https://huggingface.co/mcp"
|
34 |
self.headers = {
|
35 |
"Authorization": f"Bearer {hf_token}",
|
36 |
-
"User-Agent": "hf-mcp-client/
|
|
|
|
|
37 |
}
|
38 |
self.timeout = timedelta(seconds=timeout)
|
39 |
-
self.sse_read_timeout = timedelta(seconds=timeout * 2)
|
40 |
self.request_id_counter = 0
|
41 |
|
42 |
def _get_next_request_id(self) -> int:
|
43 |
"""Get the next request ID."""
|
44 |
self.request_id_counter += 1
|
45 |
return self.request_id_counter
|
46 |
-
|
47 |
-
async def
|
48 |
self,
|
49 |
method: str,
|
50 |
-
params: Optional[Dict[str, Any]] = None
|
51 |
-
max_retries: int = 3
|
52 |
) -> Any:
|
53 |
"""
|
54 |
-
|
55 |
-
|
56 |
-
Args:
|
57 |
-
method: The JSON-RPC method name
|
58 |
-
params: Optional parameters for the method
|
59 |
-
max_retries: Maximum number of retry attempts
|
60 |
-
|
61 |
-
Returns:
|
62 |
-
The response result or raises an exception
|
63 |
"""
|
64 |
-
last_exception = None
|
65 |
-
|
66 |
-
for attempt in range(max_retries):
|
67 |
-
try:
|
68 |
-
return await self._attempt_request(method, params)
|
69 |
-
except Exception as e:
|
70 |
-
last_exception = e
|
71 |
-
logger.warning(f"Attempt {attempt + 1} failed: {e}")
|
72 |
-
if attempt < max_retries - 1:
|
73 |
-
await asyncio.sleep(2 ** attempt) # Exponential backoff
|
74 |
-
|
75 |
-
raise last_exception
|
76 |
-
|
77 |
-
async def _attempt_request(self, method: str, params: Optional[Dict[str, Any]]) -> Any:
|
78 |
-
"""Single attempt to send request."""
|
79 |
request_id = self._get_next_request_id()
|
80 |
|
81 |
-
# Create
|
82 |
-
|
83 |
jsonrpc="2.0",
|
84 |
id=request_id,
|
85 |
method=method,
|
86 |
params=params
|
87 |
)
|
88 |
|
89 |
-
message = JSONRPCMessage(jsonrpc_request)
|
90 |
-
session_message = SessionMessage(message)
|
91 |
-
|
92 |
async with streamablehttp_client(
|
93 |
url=self.url,
|
94 |
headers=self.headers,
|
95 |
timeout=self.timeout,
|
96 |
sse_read_timeout=self.sse_read_timeout,
|
97 |
-
terminate_on_close=
|
98 |
) as (read_stream, write_stream, get_session_id):
|
99 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
try:
|
101 |
-
#
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
method="initialize",
|
106 |
-
params={
|
107 |
-
"protocolVersion": "2024-11-05",
|
108 |
-
"capabilities": {
|
109 |
-
"tools": {},
|
110 |
-
"resources": {},
|
111 |
-
"prompts": {}
|
112 |
-
},
|
113 |
-
"clientInfo": {
|
114 |
-
"name": "hf-mcp-client",
|
115 |
-
"version": "1.0.0"
|
116 |
-
}
|
117 |
-
}
|
118 |
-
)
|
119 |
-
|
120 |
-
init_message = JSONRPCMessage(init_request)
|
121 |
-
init_session_message = SessionMessage(init_message)
|
122 |
-
|
123 |
-
logger.info("Sending initialization request...")
|
124 |
-
await write_stream.send(init_session_message)
|
125 |
-
|
126 |
-
# Wait for initialization response with better error handling
|
127 |
-
init_response_received = False
|
128 |
-
timeout_counter = 0
|
129 |
-
max_iterations = 150 # Increased for Spaces environment
|
130 |
-
|
131 |
-
while not init_response_received and timeout_counter < max_iterations:
|
132 |
-
try:
|
133 |
-
# Use asyncio.wait_for to add timeout per receive operation
|
134 |
-
response = await asyncio.wait_for(
|
135 |
-
read_stream.receive(),
|
136 |
-
timeout=30.0 # 30 second timeout per receive
|
137 |
-
)
|
138 |
-
timeout_counter += 1
|
139 |
-
|
140 |
-
if isinstance(response, Exception):
|
141 |
-
logger.error(f"Received exception during init: {response}")
|
142 |
-
raise response
|
143 |
-
|
144 |
-
if isinstance(response, SessionMessage):
|
145 |
-
msg = response.message.root
|
146 |
-
if isinstance(msg, JSONRPCResponse) and msg.id == init_request.id:
|
147 |
-
logger.info("MCP client initialized successfully")
|
148 |
-
init_response_received = True
|
149 |
-
# Log the session ID if available
|
150 |
-
session_id = get_session_id()
|
151 |
-
if session_id:
|
152 |
-
logger.info(f"Session ID: {session_id}")
|
153 |
-
elif isinstance(msg, JSONRPCError) and msg.id == init_request.id:
|
154 |
-
error_msg = f"Initialization failed: {msg.error}"
|
155 |
-
logger.error(error_msg)
|
156 |
-
raise Exception(error_msg)
|
157 |
-
else:
|
158 |
-
logger.debug(f"Received other message during init: {type(msg)}")
|
159 |
-
|
160 |
-
except asyncio.TimeoutError:
|
161 |
-
logger.warning(f"Timeout waiting for response (attempt {timeout_counter})")
|
162 |
-
if timeout_counter > 10: # After 10 timeouts, give up
|
163 |
-
raise Exception("Initialization timeout: no response from server")
|
164 |
-
except Exception as e:
|
165 |
-
if "ClosedResourceError" in str(type(e)) or "StreamClosed" in str(e):
|
166 |
-
logger.error("Stream closed during initialization")
|
167 |
-
raise Exception("Connection closed during initialization")
|
168 |
-
logger.error(f"Error during initialization: {e}")
|
169 |
-
raise
|
170 |
-
|
171 |
-
if not init_response_received:
|
172 |
-
raise Exception("Initialization timeout: maximum iterations reached")
|
173 |
-
|
174 |
-
# Send initialized notification
|
175 |
-
initialized_notification = JSONRPCNotification(
|
176 |
-
jsonrpc="2.0",
|
177 |
-
method="notifications/initialized"
|
178 |
)
|
179 |
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
logger.info("Sending initialized notification...")
|
184 |
-
await write_stream.send(init_notif_session_message)
|
185 |
-
|
186 |
-
# Longer delay for Spaces environment
|
187 |
-
await asyncio.sleep(1.0)
|
188 |
-
|
189 |
-
# Now send our actual request
|
190 |
-
logger.info(f"Sending actual request: {method}")
|
191 |
-
await write_stream.send(session_message)
|
192 |
-
|
193 |
-
# Wait for the response to our request
|
194 |
-
response_received = False
|
195 |
-
timeout_counter = 0
|
196 |
-
|
197 |
-
while not response_received and timeout_counter < max_iterations:
|
198 |
-
try:
|
199 |
-
response = await asyncio.wait_for(
|
200 |
-
read_stream.receive(),
|
201 |
-
timeout=30.0
|
202 |
-
)
|
203 |
-
timeout_counter += 1
|
204 |
-
|
205 |
-
if isinstance(response, Exception):
|
206 |
-
logger.error(f"Received exception during request: {response}")
|
207 |
-
raise response
|
208 |
-
|
209 |
-
if isinstance(response, SessionMessage):
|
210 |
-
msg = response.message.root
|
211 |
-
if isinstance(msg, JSONRPCResponse) and msg.id == request_id:
|
212 |
-
logger.info(f"Request '{method}' completed successfully")
|
213 |
-
return msg.result
|
214 |
-
elif isinstance(msg, JSONRPCError) and msg.id == request_id:
|
215 |
-
error_msg = f"Request failed: {msg.error}"
|
216 |
-
logger.error(error_msg)
|
217 |
-
raise Exception(error_msg)
|
218 |
-
else:
|
219 |
-
logger.debug(f"Received other message during request: {type(msg)}")
|
220 |
-
|
221 |
-
except asyncio.TimeoutError:
|
222 |
-
logger.warning(f"Timeout waiting for request response (attempt {timeout_counter})")
|
223 |
-
if timeout_counter > 10:
|
224 |
-
raise Exception("Request timeout: no response from server")
|
225 |
-
except Exception as e:
|
226 |
-
if "ClosedResourceError" in str(type(e)) or "StreamClosed" in str(e):
|
227 |
-
logger.error("Stream closed during request processing")
|
228 |
-
raise Exception("Connection closed during request processing")
|
229 |
-
logger.error(f"Error during request processing: {e}")
|
230 |
-
raise
|
231 |
|
232 |
-
if
|
233 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
234 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
235 |
except Exception as e:
|
236 |
-
|
|
|
|
|
237 |
raise
|
238 |
-
|
239 |
-
# Ensure streams are properly closed
|
240 |
-
try:
|
241 |
-
await write_stream.aclose()
|
242 |
-
except Exception as close_error:
|
243 |
-
logger.debug(f"Error closing write stream: {close_error}")
|
244 |
-
|
245 |
async def get_all_tools(self) -> List[Dict[str, Any]]:
|
246 |
"""
|
247 |
Get all available tools from the Hugging Face MCP endpoint.
|
@@ -251,20 +206,20 @@ class HuggingFaceMCPClient:
|
|
251 |
"""
|
252 |
try:
|
253 |
logger.info("Fetching all available tools from Hugging Face MCP")
|
254 |
-
result = await self.
|
255 |
|
256 |
if isinstance(result, dict) and "tools" in result:
|
257 |
tools = result["tools"]
|
258 |
-
logger.info(f"
|
259 |
return tools
|
260 |
else:
|
261 |
-
logger.warning(f"Unexpected response format: {result}")
|
262 |
return []
|
263 |
|
264 |
except Exception as e:
|
265 |
logger.error(f"Failed to get tools: {e}")
|
266 |
raise
|
267 |
-
|
268 |
async def call_tool(self, tool_name: str, args: Dict[str, Any]) -> Any:
|
269 |
"""
|
270 |
Call a specific tool with the given arguments.
|
@@ -284,7 +239,7 @@ class HuggingFaceMCPClient:
|
|
284 |
"arguments": args
|
285 |
}
|
286 |
|
287 |
-
result = await self.
|
288 |
logger.info(f"Tool '{tool_name}' executed successfully")
|
289 |
return result
|
290 |
|
@@ -293,92 +248,351 @@ class HuggingFaceMCPClient:
|
|
293 |
raise
|
294 |
|
295 |
|
296 |
-
|
297 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
298 |
"""
|
299 |
-
Get all available tools
|
300 |
|
301 |
Args:
|
302 |
hf_token: Hugging Face API token
|
|
|
303 |
|
304 |
Returns:
|
305 |
List of tool definitions
|
306 |
"""
|
307 |
-
|
308 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
309 |
|
310 |
|
311 |
-
async def
|
|
|
|
|
|
|
|
|
|
|
312 |
"""
|
313 |
-
Call a specific Hugging Face MCP tool.
|
314 |
|
315 |
Args:
|
316 |
hf_token: Hugging Face API token
|
317 |
tool_name: Name of the tool to call
|
318 |
args: Arguments to pass to the tool
|
|
|
319 |
|
320 |
Returns:
|
321 |
The tool's response
|
322 |
"""
|
323 |
-
|
324 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
325 |
|
326 |
|
327 |
-
#
|
328 |
-
async def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
329 |
"""
|
330 |
-
|
331 |
|
332 |
Args:
|
333 |
hf_token: Hugging Face API token
|
334 |
|
335 |
Returns:
|
336 |
-
|
337 |
"""
|
338 |
diagnostics = {
|
339 |
"environment": "huggingface_spaces" if os.getenv("SPACE_ID") else "local",
|
340 |
"space_id": os.getenv("SPACE_ID"),
|
|
|
341 |
"token_length": len(hf_token) if hf_token else 0,
|
342 |
"has_token": bool(hf_token),
|
343 |
-
"
|
344 |
-
|
345 |
-
|
346 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
347 |
}
|
348 |
|
|
|
349 |
try:
|
350 |
-
# Test basic connection
|
351 |
-
from mcp.client.streamable_http import streamablehttp_client
|
352 |
-
|
353 |
-
headers = {
|
354 |
-
"Authorization": f"Bearer {hf_token}",
|
355 |
-
"User-Agent": "hf-mcp-diagnostic/1.0.0"
|
356 |
-
}
|
357 |
-
|
358 |
async with streamablehttp_client(
|
359 |
url="https://huggingface.co/mcp",
|
360 |
-
headers=
|
361 |
-
timeout=timedelta(seconds=
|
362 |
-
terminate_on_close=
|
363 |
) as (read_stream, write_stream, get_session_id):
|
364 |
-
diagnostics["
|
365 |
logger.info("Basic connection test passed")
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
373 |
diagnostics["tool_count"] = len(tools)
|
374 |
-
|
375 |
-
|
376 |
-
|
377 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
378 |
|
379 |
-
|
380 |
-
|
381 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
382 |
|
383 |
-
return diagnostics
|
384 |
-
|
|
|
1 |
+
"""
|
2 |
+
Robust Hugging Face MCP Client - Optimized for HF Spaces
|
3 |
+
|
4 |
+
This module provides a robust client for interacting with Hugging Face's MCP endpoint
|
5 |
+
with better error handling, TaskGroup avoidance, and compatibility for Hugging Face Spaces.
|
6 |
+
"""
|
7 |
+
|
8 |
import asyncio
|
9 |
import json
|
10 |
import logging
|
11 |
import os
|
12 |
+
from typing import Any, Dict, List, Optional, Union
|
13 |
from datetime import timedelta
|
14 |
+
from contextlib import asynccontextmanager
|
15 |
|
16 |
from mcp.shared.message import SessionMessage
|
17 |
from mcp.types import (
|
|
|
26 |
logger = logging.getLogger(__name__)
|
27 |
|
28 |
|
29 |
+
class RobustHFMCPClient:
|
30 |
+
"""Robust client for interacting with Hugging Face MCP endpoint optimized for Spaces."""
|
31 |
|
32 |
+
def __init__(self, hf_token: str, timeout: int = 120):
|
33 |
"""
|
34 |
+
Initialize the Robust Hugging Face MCP client.
|
35 |
|
36 |
Args:
|
37 |
hf_token: Hugging Face API token
|
38 |
+
timeout: Timeout in seconds for HTTP requests
|
39 |
"""
|
40 |
self.hf_token = hf_token
|
41 |
self.url = "https://huggingface.co/mcp"
|
42 |
self.headers = {
|
43 |
"Authorization": f"Bearer {hf_token}",
|
44 |
+
"User-Agent": "robust-hf-mcp-client/2.0.0",
|
45 |
+
"Accept": "application/json, text/event-stream",
|
46 |
+
"Content-Type": "application/json"
|
47 |
}
|
48 |
self.timeout = timedelta(seconds=timeout)
|
49 |
+
self.sse_read_timeout = timedelta(seconds=timeout * 2)
|
50 |
self.request_id_counter = 0
|
51 |
|
52 |
def _get_next_request_id(self) -> int:
|
53 |
"""Get the next request ID."""
|
54 |
self.request_id_counter += 1
|
55 |
return self.request_id_counter
|
56 |
+
|
57 |
+
async def _execute_single_request_session(
|
58 |
self,
|
59 |
method: str,
|
60 |
+
params: Optional[Dict[str, Any]] = None
|
|
|
61 |
) -> Any:
|
62 |
"""
|
63 |
+
Execute a complete MCP session for a single request.
|
64 |
+
This avoids TaskGroup issues by handling everything in sequence.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
request_id = self._get_next_request_id()
|
67 |
|
68 |
+
# Create the main request
|
69 |
+
main_request = JSONRPCRequest(
|
70 |
jsonrpc="2.0",
|
71 |
id=request_id,
|
72 |
method=method,
|
73 |
params=params
|
74 |
)
|
75 |
|
|
|
|
|
|
|
76 |
async with streamablehttp_client(
|
77 |
url=self.url,
|
78 |
headers=self.headers,
|
79 |
timeout=self.timeout,
|
80 |
sse_read_timeout=self.sse_read_timeout,
|
81 |
+
terminate_on_close=False # Avoid TaskGroup cleanup issues
|
82 |
) as (read_stream, write_stream, get_session_id):
|
83 |
|
84 |
+
# Step 1: Initialize the session
|
85 |
+
logger.info("Starting MCP session initialization...")
|
86 |
+
await self._initialize_session(read_stream, write_stream)
|
87 |
+
|
88 |
+
# Step 2: Send the main request
|
89 |
+
logger.info(f"Sending main request: {method}")
|
90 |
+
main_message = JSONRPCMessage(main_request)
|
91 |
+
main_session_message = SessionMessage(main_message)
|
92 |
+
await write_stream.send(main_session_message)
|
93 |
+
|
94 |
+
# Step 3: Wait for the response
|
95 |
+
logger.info("Waiting for main request response...")
|
96 |
+
response = await self._wait_for_response(read_stream, request_id, timeout=90)
|
97 |
+
|
98 |
+
return response
|
99 |
+
|
100 |
+
async def _initialize_session(self, read_stream, write_stream) -> None:
|
101 |
+
"""Initialize the MCP session with proper handshake."""
|
102 |
+
init_request_id = self._get_next_request_id()
|
103 |
+
|
104 |
+
# Send initialize request
|
105 |
+
init_request = JSONRPCRequest(
|
106 |
+
jsonrpc="2.0",
|
107 |
+
id=init_request_id,
|
108 |
+
method="initialize",
|
109 |
+
params={
|
110 |
+
"protocolVersion": "2024-11-05",
|
111 |
+
"capabilities": {
|
112 |
+
"tools": {},
|
113 |
+
"resources": {},
|
114 |
+
"prompts": {}
|
115 |
+
},
|
116 |
+
"clientInfo": {
|
117 |
+
"name": "robust-hf-mcp-client",
|
118 |
+
"version": "2.0.0"
|
119 |
+
}
|
120 |
+
}
|
121 |
+
)
|
122 |
+
|
123 |
+
init_message = JSONRPCMessage(init_request)
|
124 |
+
init_session_message = SessionMessage(init_message)
|
125 |
+
|
126 |
+
await write_stream.send(init_session_message)
|
127 |
+
|
128 |
+
# Wait for initialization response
|
129 |
+
init_response = await self._wait_for_response(read_stream, init_request_id, timeout=60)
|
130 |
+
logger.info("MCP session initialized successfully")
|
131 |
+
|
132 |
+
# Send initialized notification
|
133 |
+
initialized_notification = JSONRPCNotification(
|
134 |
+
jsonrpc="2.0",
|
135 |
+
method="notifications/initialized"
|
136 |
+
)
|
137 |
+
|
138 |
+
init_notif_message = JSONRPCMessage(initialized_notification)
|
139 |
+
init_notif_session_message = SessionMessage(init_notif_message)
|
140 |
+
|
141 |
+
await write_stream.send(init_notif_session_message)
|
142 |
+
|
143 |
+
# Give the server time to process the notification
|
144 |
+
await asyncio.sleep(1.0)
|
145 |
+
|
146 |
+
async def _wait_for_response(
|
147 |
+
self,
|
148 |
+
read_stream,
|
149 |
+
expected_id: int,
|
150 |
+
timeout: int = 60
|
151 |
+
) -> Any:
|
152 |
+
"""
|
153 |
+
Wait for a specific response by ID with timeout handling.
|
154 |
+
"""
|
155 |
+
start_time = asyncio.get_event_loop().time()
|
156 |
+
|
157 |
+
while True:
|
158 |
+
current_time = asyncio.get_event_loop().time()
|
159 |
+
if current_time - start_time > timeout:
|
160 |
+
raise asyncio.TimeoutError(f"Timeout waiting for response to request {expected_id}")
|
161 |
+
|
162 |
try:
|
163 |
+
# Use a shorter timeout for each receive to avoid hanging
|
164 |
+
response = await asyncio.wait_for(
|
165 |
+
read_stream.receive(),
|
166 |
+
timeout=10.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
)
|
168 |
|
169 |
+
if isinstance(response, Exception):
|
170 |
+
logger.error(f"Received exception in stream: {response}")
|
171 |
+
raise response
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
172 |
|
173 |
+
if isinstance(response, SessionMessage):
|
174 |
+
msg_root = response.message.root
|
175 |
+
|
176 |
+
if isinstance(msg_root, JSONRPCResponse) and msg_root.id == expected_id:
|
177 |
+
logger.info(f"Received successful response for request {expected_id}")
|
178 |
+
return msg_root.result
|
179 |
+
|
180 |
+
elif isinstance(msg_root, JSONRPCError) and msg_root.id == expected_id:
|
181 |
+
error_msg = f"Server error for request {expected_id}: {msg_root.error}"
|
182 |
+
logger.error(error_msg)
|
183 |
+
raise Exception(error_msg)
|
184 |
|
185 |
+
else:
|
186 |
+
# Log unexpected messages but continue waiting
|
187 |
+
logger.debug(f"Received unexpected message type: {type(msg_root)} with ID: {getattr(msg_root, 'id', 'N/A')}")
|
188 |
+
continue
|
189 |
+
|
190 |
+
except asyncio.TimeoutError:
|
191 |
+
# Continue the outer loop to check the overall timeout
|
192 |
+
logger.debug("Receive timeout, continuing to wait...")
|
193 |
+
continue
|
194 |
except Exception as e:
|
195 |
+
if "ClosedResourceError" in str(type(e)) or "StreamClosed" in str(e):
|
196 |
+
raise Exception("Connection closed while waiting for response")
|
197 |
+
logger.error(f"Error while waiting for response: {e}")
|
198 |
raise
|
199 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
200 |
async def get_all_tools(self) -> List[Dict[str, Any]]:
|
201 |
"""
|
202 |
Get all available tools from the Hugging Face MCP endpoint.
|
|
|
206 |
"""
|
207 |
try:
|
208 |
logger.info("Fetching all available tools from Hugging Face MCP")
|
209 |
+
result = await self._execute_single_request_session("tools/list")
|
210 |
|
211 |
if isinstance(result, dict) and "tools" in result:
|
212 |
tools = result["tools"]
|
213 |
+
logger.info(f"Successfully fetched {len(tools)} tools")
|
214 |
return tools
|
215 |
else:
|
216 |
+
logger.warning(f"Unexpected response format for tools/list: {result}")
|
217 |
return []
|
218 |
|
219 |
except Exception as e:
|
220 |
logger.error(f"Failed to get tools: {e}")
|
221 |
raise
|
222 |
+
|
223 |
async def call_tool(self, tool_name: str, args: Dict[str, Any]) -> Any:
|
224 |
"""
|
225 |
Call a specific tool with the given arguments.
|
|
|
239 |
"arguments": args
|
240 |
}
|
241 |
|
242 |
+
result = await self._execute_single_request_session("tools/call", params)
|
243 |
logger.info(f"Tool '{tool_name}' executed successfully")
|
244 |
return result
|
245 |
|
|
|
248 |
raise
|
249 |
|
250 |
|
251 |
+
class SimplifiedHFMCPClient:
|
252 |
+
"""Ultra-simplified client that avoids all TaskGroup usage."""
|
253 |
+
|
254 |
+
def __init__(self, hf_token: str, timeout: int = 90):
|
255 |
+
self.hf_token = hf_token
|
256 |
+
self.timeout = timeout
|
257 |
+
self.headers = {
|
258 |
+
"Authorization": f"Bearer {hf_token}",
|
259 |
+
"User-Agent": "simplified-hf-mcp-client/1.0.0"
|
260 |
+
}
|
261 |
+
self.request_counter = 0
|
262 |
+
|
263 |
+
def _next_id(self) -> int:
|
264 |
+
self.request_counter += 1
|
265 |
+
return self.request_counter
|
266 |
+
|
267 |
+
async def _simple_mcp_call(self, method: str, params: Optional[Dict[str, Any]] = None) -> Any:
|
268 |
+
"""Make a simple MCP call without complex async patterns."""
|
269 |
+
|
270 |
+
async with streamablehttp_client(
|
271 |
+
url="https://huggingface.co/mcp",
|
272 |
+
headers=self.headers,
|
273 |
+
timeout=timedelta(seconds=self.timeout),
|
274 |
+
sse_read_timeout=timedelta(seconds=self.timeout * 2),
|
275 |
+
terminate_on_close=False
|
276 |
+
) as (read_stream, write_stream, get_session_id):
|
277 |
+
|
278 |
+
responses = {}
|
279 |
+
|
280 |
+
# Simple message handler
|
281 |
+
async def collect_responses():
|
282 |
+
try:
|
283 |
+
async for message in read_stream:
|
284 |
+
if isinstance(message, Exception):
|
285 |
+
responses['error'] = message
|
286 |
+
break
|
287 |
+
elif isinstance(message, SessionMessage):
|
288 |
+
msg_root = message.message.root
|
289 |
+
if hasattr(msg_root, 'id') and msg_root.id is not None:
|
290 |
+
responses[msg_root.id] = msg_root
|
291 |
+
except Exception as e:
|
292 |
+
responses['error'] = e
|
293 |
+
|
294 |
+
# Start response collector
|
295 |
+
collector_task = asyncio.create_task(collect_responses())
|
296 |
+
|
297 |
+
try:
|
298 |
+
# Step 1: Initialize
|
299 |
+
init_id = self._next_id()
|
300 |
+
init_req = JSONRPCRequest(
|
301 |
+
jsonrpc="2.0",
|
302 |
+
id=init_id,
|
303 |
+
method="initialize",
|
304 |
+
params={
|
305 |
+
"protocolVersion": "2024-11-05",
|
306 |
+
"capabilities": {"tools": {}},
|
307 |
+
"clientInfo": {"name": "simple-hf-mcp", "version": "1.0.0"}
|
308 |
+
}
|
309 |
+
)
|
310 |
+
|
311 |
+
await write_stream.send(SessionMessage(JSONRPCMessage(init_req)))
|
312 |
+
|
313 |
+
# Wait for init response
|
314 |
+
for _ in range(300): # 30 seconds max
|
315 |
+
if init_id in responses:
|
316 |
+
break
|
317 |
+
if 'error' in responses:
|
318 |
+
raise responses['error']
|
319 |
+
await asyncio.sleep(0.1)
|
320 |
+
|
321 |
+
if init_id not in responses:
|
322 |
+
raise Exception("Initialization timeout")
|
323 |
+
|
324 |
+
# Step 2: Send initialized notification
|
325 |
+
notif = JSONRPCNotification(
|
326 |
+
jsonrpc="2.0",
|
327 |
+
method="notifications/initialized"
|
328 |
+
)
|
329 |
+
await write_stream.send(SessionMessage(JSONRPCMessage(notif)))
|
330 |
+
await asyncio.sleep(0.5)
|
331 |
+
|
332 |
+
# Step 3: Send main request
|
333 |
+
main_id = self._next_id()
|
334 |
+
main_req = JSONRPCRequest(
|
335 |
+
jsonrpc="2.0",
|
336 |
+
id=main_id,
|
337 |
+
method=method,
|
338 |
+
params=params
|
339 |
+
)
|
340 |
+
|
341 |
+
await write_stream.send(SessionMessage(JSONRPCMessage(main_req)))
|
342 |
+
|
343 |
+
# Wait for main response
|
344 |
+
for _ in range(600): # 60 seconds max
|
345 |
+
if main_id in responses:
|
346 |
+
break
|
347 |
+
if 'error' in responses:
|
348 |
+
raise responses['error']
|
349 |
+
await asyncio.sleep(0.1)
|
350 |
+
|
351 |
+
if main_id not in responses:
|
352 |
+
raise Exception("Main request timeout")
|
353 |
+
|
354 |
+
result = responses[main_id]
|
355 |
+
if isinstance(result, JSONRPCResponse):
|
356 |
+
return result.result
|
357 |
+
elif isinstance(result, JSONRPCError):
|
358 |
+
raise Exception(f"Server error: {result.error}")
|
359 |
+
else:
|
360 |
+
raise Exception(f"Unexpected response type: {type(result)}")
|
361 |
+
|
362 |
+
finally:
|
363 |
+
collector_task.cancel()
|
364 |
+
try:
|
365 |
+
await collector_task
|
366 |
+
except asyncio.CancelledError:
|
367 |
+
pass
|
368 |
+
|
369 |
+
async def get_tools(self) -> List[Dict[str, Any]]:
|
370 |
+
"""Get all available tools."""
|
371 |
+
result = await self._simple_mcp_call("tools/list")
|
372 |
+
if isinstance(result, dict) and "tools" in result:
|
373 |
+
return result["tools"]
|
374 |
+
return []
|
375 |
+
|
376 |
+
async def call_tool(self, tool_name: str, args: Dict[str, Any]) -> Any:
|
377 |
+
"""Call a specific tool."""
|
378 |
+
params = {
|
379 |
+
"name": tool_name,
|
380 |
+
"arguments": args
|
381 |
+
}
|
382 |
+
return await self._simple_mcp_call("tools/call", params)
|
383 |
+
|
384 |
+
|
385 |
+
# Robust convenience functions
|
386 |
+
async def get_hf_tools_robust(hf_token: str, max_retries: int = 3) -> List[Dict[str, Any]]:
|
387 |
"""
|
388 |
+
Get all available tools with multiple fallback strategies.
|
389 |
|
390 |
Args:
|
391 |
hf_token: Hugging Face API token
|
392 |
+
max_retries: Maximum retry attempts per method
|
393 |
|
394 |
Returns:
|
395 |
List of tool definitions
|
396 |
"""
|
397 |
+
last_error = None
|
398 |
+
|
399 |
+
# Strategy 1: Try the robust client
|
400 |
+
for attempt in range(max_retries):
|
401 |
+
try:
|
402 |
+
logger.info(f"Trying robust client (attempt {attempt + 1})")
|
403 |
+
client = RobustHFMCPClient(hf_token, timeout=90)
|
404 |
+
tools = await client.get_all_tools()
|
405 |
+
logger.info(f"Robust client succeeded with {len(tools)} tools")
|
406 |
+
return tools
|
407 |
+
except Exception as e:
|
408 |
+
last_error = e
|
409 |
+
logger.warning(f"Robust client attempt {attempt + 1} failed: {e}")
|
410 |
+
if attempt < max_retries - 1:
|
411 |
+
await asyncio.sleep(2 ** attempt) # Exponential backoff
|
412 |
+
|
413 |
+
# Strategy 2: Try the simplified client
|
414 |
+
for attempt in range(max_retries):
|
415 |
+
try:
|
416 |
+
logger.info(f"Trying simplified client (attempt {attempt + 1})")
|
417 |
+
client = SimplifiedHFMCPClient(hf_token, timeout=120)
|
418 |
+
tools = await client.get_tools()
|
419 |
+
logger.info(f"Simplified client succeeded with {len(tools)} tools")
|
420 |
+
return tools
|
421 |
+
except Exception as e:
|
422 |
+
last_error = e
|
423 |
+
logger.warning(f"Simplified client attempt {attempt + 1} failed: {e}")
|
424 |
+
if attempt < max_retries - 1:
|
425 |
+
await asyncio.sleep(2 ** attempt)
|
426 |
+
|
427 |
+
# If all strategies fail
|
428 |
+
raise Exception(f"All connection strategies failed. Last error: {last_error}")
|
429 |
|
430 |
|
431 |
+
async def call_hf_tool_robust(
|
432 |
+
hf_token: str,
|
433 |
+
tool_name: str,
|
434 |
+
args: Dict[str, Any],
|
435 |
+
max_retries: int = 3
|
436 |
+
) -> Any:
|
437 |
"""
|
438 |
+
Call a specific Hugging Face MCP tool with multiple fallback strategies.
|
439 |
|
440 |
Args:
|
441 |
hf_token: Hugging Face API token
|
442 |
tool_name: Name of the tool to call
|
443 |
args: Arguments to pass to the tool
|
444 |
+
max_retries: Maximum retry attempts per method
|
445 |
|
446 |
Returns:
|
447 |
The tool's response
|
448 |
"""
|
449 |
+
last_error = None
|
450 |
+
|
451 |
+
# Strategy 1: Try the robust client
|
452 |
+
for attempt in range(max_retries):
|
453 |
+
try:
|
454 |
+
logger.info(f"Trying robust client for tool call (attempt {attempt + 1})")
|
455 |
+
client = RobustHFMCPClient(hf_token, timeout=120)
|
456 |
+
result = await client.call_tool(tool_name, args)
|
457 |
+
logger.info(f"Robust client tool call succeeded")
|
458 |
+
return result
|
459 |
+
except Exception as e:
|
460 |
+
last_error = e
|
461 |
+
logger.warning(f"Robust client tool call attempt {attempt + 1} failed: {e}")
|
462 |
+
if attempt < max_retries - 1:
|
463 |
+
await asyncio.sleep(2 ** attempt)
|
464 |
+
|
465 |
+
# Strategy 2: Try the simplified client
|
466 |
+
for attempt in range(max_retries):
|
467 |
+
try:
|
468 |
+
logger.info(f"Trying simplified client for tool call (attempt {attempt + 1})")
|
469 |
+
client = SimplifiedHFMCPClient(hf_token, timeout=150)
|
470 |
+
result = await client.call_tool(tool_name, args)
|
471 |
+
logger.info(f"Simplified client tool call succeeded")
|
472 |
+
return result
|
473 |
+
except Exception as e:
|
474 |
+
last_error = e
|
475 |
+
logger.warning(f"Simplified client tool call attempt {attempt + 1} failed: {e}")
|
476 |
+
if attempt < max_retries - 1:
|
477 |
+
await asyncio.sleep(2 ** attempt)
|
478 |
+
|
479 |
+
# If all strategies fail
|
480 |
+
raise Exception(f"All tool call strategies failed. Last error: {last_error}")
|
481 |
|
482 |
|
483 |
+
# Legacy compatibility functions
|
484 |
+
async def get_hf_tools(hf_token: str) -> List[Dict[str, Any]]:
|
485 |
+
"""Legacy function - now uses robust implementation."""
|
486 |
+
return await get_hf_tools_robust(hf_token)
|
487 |
+
|
488 |
+
|
489 |
+
async def call_hf_tool(hf_token: str, tool_name: str, args: Dict[str, Any]) -> Any:
|
490 |
+
"""Legacy function - now uses robust implementation."""
|
491 |
+
return await call_hf_tool_robust(hf_token, tool_name, args)
|
492 |
+
|
493 |
+
|
494 |
+
# Enhanced diagnostics
|
495 |
+
async def diagnose_connection_advanced(hf_token: str) -> Dict[str, Any]:
|
496 |
"""
|
497 |
+
Advanced connection diagnostics with multiple test scenarios.
|
498 |
|
499 |
Args:
|
500 |
hf_token: Hugging Face API token
|
501 |
|
502 |
Returns:
|
503 |
+
Comprehensive diagnostic information
|
504 |
"""
|
505 |
diagnostics = {
|
506 |
"environment": "huggingface_spaces" if os.getenv("SPACE_ID") else "local",
|
507 |
"space_id": os.getenv("SPACE_ID"),
|
508 |
+
"python_version": os.sys.version,
|
509 |
"token_length": len(hf_token) if hf_token else 0,
|
510 |
"has_token": bool(hf_token),
|
511 |
+
"tests": {
|
512 |
+
"basic_connection": False,
|
513 |
+
"robust_client": False,
|
514 |
+
"simplified_client": False,
|
515 |
+
"tools_fetch": False,
|
516 |
+
"tool_call_test": False
|
517 |
+
},
|
518 |
+
"errors": {},
|
519 |
+
"tool_count": 0,
|
520 |
+
"sample_tools": []
|
521 |
}
|
522 |
|
523 |
+
# Test 1: Basic connection
|
524 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
525 |
async with streamablehttp_client(
|
526 |
url="https://huggingface.co/mcp",
|
527 |
+
headers={"Authorization": f"Bearer {hf_token}"},
|
528 |
+
timeout=timedelta(seconds=10),
|
529 |
+
terminate_on_close=False
|
530 |
) as (read_stream, write_stream, get_session_id):
|
531 |
+
diagnostics["tests"]["basic_connection"] = True
|
532 |
logger.info("Basic connection test passed")
|
533 |
+
except Exception as e:
|
534 |
+
diagnostics["errors"]["basic_connection"] = str(e)
|
535 |
+
logger.error(f"Basic connection test failed: {e}")
|
536 |
+
|
537 |
+
# Test 2: Robust client
|
538 |
+
if diagnostics["tests"]["basic_connection"]:
|
539 |
+
try:
|
540 |
+
client = RobustHFMCPClient(hf_token, timeout=60)
|
541 |
+
tools = await client.get_all_tools()
|
542 |
+
diagnostics["tests"]["robust_client"] = True
|
543 |
+
diagnostics["tests"]["tools_fetch"] = True
|
544 |
+
diagnostics["tool_count"] = len(tools)
|
545 |
+
diagnostics["sample_tools"] = [
|
546 |
+
{"name": tool.get("name"), "description": tool.get("description", "")[:100]}
|
547 |
+
for tool in tools[:3]
|
548 |
+
]
|
549 |
+
logger.info(f"Robust client test passed - {len(tools)} tools")
|
550 |
+
except Exception as e:
|
551 |
+
diagnostics["errors"]["robust_client"] = str(e)
|
552 |
+
logger.error(f"Robust client test failed: {e}")
|
553 |
+
|
554 |
+
# Test 3: Simplified client
|
555 |
+
if not diagnostics["tests"]["robust_client"]:
|
556 |
+
try:
|
557 |
+
client = SimplifiedHFMCPClient(hf_token, timeout=90)
|
558 |
+
tools = await client.get_tools()
|
559 |
+
diagnostics["tests"]["simplified_client"] = True
|
560 |
+
if not diagnostics["tests"]["tools_fetch"]:
|
561 |
+
diagnostics["tests"]["tools_fetch"] = True
|
562 |
diagnostics["tool_count"] = len(tools)
|
563 |
+
diagnostics["sample_tools"] = [
|
564 |
+
{"name": tool.get("name"), "description": tool.get("description", "")[:100]}
|
565 |
+
for tool in tools[:3]
|
566 |
+
]
|
567 |
+
logger.info(f"Simplified client test passed - {len(tools)} tools")
|
568 |
+
except Exception as e:
|
569 |
+
diagnostics["errors"]["simplified_client"] = str(e)
|
570 |
+
logger.error(f"Simplified client test failed: {e}")
|
571 |
+
|
572 |
+
# Test 4: Tool call (if we have tools)
|
573 |
+
if diagnostics["tests"]["tools_fetch"] and diagnostics["sample_tools"]:
|
574 |
+
try:
|
575 |
+
# Try to call a simple tool if available
|
576 |
+
sample_tool_name = diagnostics["sample_tools"][0]["name"]
|
577 |
+
if sample_tool_name:
|
578 |
+
# Use the working client
|
579 |
+
if diagnostics["tests"]["robust_client"]:
|
580 |
+
client = RobustHFMCPClient(hf_token, timeout=60)
|
581 |
+
else:
|
582 |
+
client = SimplifiedHFMCPClient(hf_token, timeout=90)
|
583 |
|
584 |
+
# Try with empty args first (many tools accept this)
|
585 |
+
try:
|
586 |
+
result = await client.call_tool(sample_tool_name, {})
|
587 |
+
diagnostics["tests"]["tool_call_test"] = True
|
588 |
+
logger.info(f"Tool call test passed with {sample_tool_name}")
|
589 |
+
except Exception as tool_error:
|
590 |
+
# Tool call failed but that might be due to wrong args
|
591 |
+
diagnostics["errors"]["tool_call_test"] = f"Tool call failed (might need args): {str(tool_error)}"
|
592 |
+
logger.warning(f"Tool call test failed: {tool_error}")
|
593 |
+
|
594 |
+
except Exception as e:
|
595 |
+
diagnostics["errors"]["tool_call_test"] = str(e)
|
596 |
+
logger.error(f"Tool call test setup failed: {e}")
|
597 |
|
598 |
+
return diagnostics
|
|