Adding all project files
This commit is contained in:
parent
6c9e127bdc
commit
cd4316ad0f
42289 changed files with 8009643 additions and 0 deletions
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
|
|
@ -0,0 +1,88 @@
|
|||
import asyncio
|
||||
import sys
|
||||
from functools import partial
|
||||
|
||||
import typer
|
||||
|
||||
|
||||
def _patch_anyio_open_process():
|
||||
"""
|
||||
Patch anyio.open_process to allow detached processes on Windows and Unix-like systems.
|
||||
|
||||
This is necessary to prevent the MCP client from being interrupted by Ctrl+C when running in the CLI.
|
||||
"""
|
||||
import subprocess
|
||||
|
||||
import anyio
|
||||
|
||||
if getattr(anyio, "_tiny_agents_patched", False):
|
||||
return
|
||||
anyio._tiny_agents_patched = True
|
||||
|
||||
original_open_process = anyio.open_process
|
||||
|
||||
if sys.platform == "win32":
|
||||
# On Windows, we need to set the creation flags to create a new process group
|
||||
|
||||
async def open_process_in_new_group(*args, **kwargs):
|
||||
"""
|
||||
Wrapper for open_process to handle Windows-specific process creation flags.
|
||||
"""
|
||||
# Ensure we pass the creation flags for Windows
|
||||
kwargs.setdefault("creationflags", subprocess.CREATE_NEW_PROCESS_GROUP)
|
||||
return await original_open_process(*args, **kwargs)
|
||||
|
||||
anyio.open_process = open_process_in_new_group
|
||||
else:
|
||||
# For Unix-like systems, we can use setsid to create a new session
|
||||
async def open_process_in_new_group(*args, **kwargs):
|
||||
"""
|
||||
Wrapper for open_process to handle Unix-like systems with start_new_session=True.
|
||||
"""
|
||||
kwargs.setdefault("start_new_session", True)
|
||||
return await original_open_process(*args, **kwargs)
|
||||
|
||||
anyio.open_process = open_process_in_new_group
|
||||
|
||||
|
||||
async def _async_prompt(exit_event: asyncio.Event, prompt: str = "» ") -> str:
|
||||
"""
|
||||
Asynchronous prompt function that reads input from stdin without blocking.
|
||||
|
||||
This function is designed to work in an asynchronous context, allowing the event loop to gracefully stop it (e.g. on Ctrl+C).
|
||||
|
||||
Alternatively, we could use https://github.com/vxgmichel/aioconsole but that would be an additional dependency.
|
||||
"""
|
||||
loop = asyncio.get_event_loop()
|
||||
|
||||
if sys.platform == "win32":
|
||||
# Windows: Use run_in_executor to avoid blocking the event loop
|
||||
# Degraded solution: this is not ideal as user will have to CTRL+C once more to stop the prompt (and it'll not be graceful)
|
||||
return await loop.run_in_executor(None, partial(typer.prompt, prompt, prompt_suffix=" "))
|
||||
else:
|
||||
# UNIX-like: Use loop.add_reader for non-blocking stdin read
|
||||
future = loop.create_future()
|
||||
|
||||
def on_input():
|
||||
line = sys.stdin.readline()
|
||||
loop.remove_reader(sys.stdin)
|
||||
future.set_result(line)
|
||||
|
||||
print(prompt, end=" ", flush=True)
|
||||
loop.add_reader(sys.stdin, on_input) # not supported on Windows
|
||||
|
||||
# Wait for user input or exit event
|
||||
# Wait until either the user hits enter or exit_event is set
|
||||
exit_task = asyncio.create_task(exit_event.wait())
|
||||
await asyncio.wait(
|
||||
[future, exit_task],
|
||||
return_when=asyncio.FIRST_COMPLETED,
|
||||
)
|
||||
|
||||
# Check which one has been triggered
|
||||
if exit_event.is_set():
|
||||
future.cancel()
|
||||
return ""
|
||||
|
||||
line = await future
|
||||
return line.strip()
|
||||
103
venv/Lib/site-packages/huggingface_hub/inference/_mcp/agent.py
Normal file
103
venv/Lib/site-packages/huggingface_hub/inference/_mcp/agent.py
Normal file
|
|
@ -0,0 +1,103 @@
|
|||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
from typing import AsyncGenerator, Dict, Iterable, List, Optional, Union
|
||||
|
||||
from huggingface_hub import ChatCompletionInputMessage, ChatCompletionStreamOutput, MCPClient
|
||||
|
||||
from .._providers import PROVIDER_OR_POLICY_T
|
||||
from .constants import DEFAULT_SYSTEM_PROMPT, EXIT_LOOP_TOOLS, MAX_NUM_TURNS
|
||||
from .types import ServerConfig
|
||||
|
||||
|
||||
class Agent(MCPClient):
|
||||
"""
|
||||
Implementation of a Simple Agent, which is a simple while loop built right on top of an [`MCPClient`].
|
||||
|
||||
<Tip warning={true}>
|
||||
|
||||
This class is experimental and might be subject to breaking changes in the future without prior notice.
|
||||
|
||||
</Tip>
|
||||
|
||||
Args:
|
||||
model (`str`, *optional*):
|
||||
The model to run inference with. Can be a model id hosted on the Hugging Face Hub, e.g. `meta-llama/Meta-Llama-3-8B-Instruct`
|
||||
or a URL to a deployed Inference Endpoint or other local or remote endpoint.
|
||||
servers (`Iterable[Dict]`):
|
||||
MCP servers to connect to. Each server is a dictionary containing a `type` key and a `config` key. The `type` key can be `"stdio"` or `"sse"`, and the `config` key is a dictionary of arguments for the server.
|
||||
provider (`str`, *optional*):
|
||||
Name of the provider to use for inference. Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
|
||||
If model is a URL or `base_url` is passed, then `provider` is not used.
|
||||
base_url (`str`, *optional*):
|
||||
The base URL to run inference. Defaults to None.
|
||||
api_key (`str`, *optional*):
|
||||
Token to use for authentication. Will default to the locally Hugging Face saved token if not provided. You can also use your own provider API key to interact directly with the provider's service.
|
||||
prompt (`str`, *optional*):
|
||||
The system prompt to use for the agent. Defaults to the default system prompt in `constants.py`.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
model: Optional[str] = None,
|
||||
servers: Iterable[ServerConfig],
|
||||
provider: Optional[PROVIDER_OR_POLICY_T] = None,
|
||||
base_url: Optional[str] = None,
|
||||
api_key: Optional[str] = None,
|
||||
prompt: Optional[str] = None,
|
||||
):
|
||||
super().__init__(model=model, provider=provider, base_url=base_url, api_key=api_key)
|
||||
self._servers_cfg = list(servers)
|
||||
self.messages: List[Union[Dict, ChatCompletionInputMessage]] = [
|
||||
{"role": "system", "content": prompt or DEFAULT_SYSTEM_PROMPT}
|
||||
]
|
||||
|
||||
async def load_tools(self) -> None:
|
||||
for cfg in self._servers_cfg:
|
||||
await self.add_mcp_server(**cfg)
|
||||
|
||||
async def run(
|
||||
self,
|
||||
user_input: str,
|
||||
*,
|
||||
abort_event: Optional[asyncio.Event] = None,
|
||||
) -> AsyncGenerator[Union[ChatCompletionStreamOutput, ChatCompletionInputMessage], None]:
|
||||
"""
|
||||
Run the agent with the given user input.
|
||||
|
||||
Args:
|
||||
user_input (`str`):
|
||||
The user input to run the agent with.
|
||||
abort_event (`asyncio.Event`, *optional*):
|
||||
An event that can be used to abort the agent. If the event is set, the agent will stop running.
|
||||
"""
|
||||
self.messages.append({"role": "user", "content": user_input})
|
||||
|
||||
num_turns: int = 0
|
||||
next_turn_should_call_tools = True
|
||||
|
||||
while True:
|
||||
if abort_event and abort_event.is_set():
|
||||
return
|
||||
|
||||
async for item in self.process_single_turn_with_tools(
|
||||
self.messages,
|
||||
exit_loop_tools=EXIT_LOOP_TOOLS,
|
||||
exit_if_first_chunk_no_tool=(num_turns > 0 and next_turn_should_call_tools),
|
||||
):
|
||||
yield item
|
||||
|
||||
num_turns += 1
|
||||
last = self.messages[-1]
|
||||
|
||||
if last.get("role") == "tool" and last.get("name") in {t.function.name for t in EXIT_LOOP_TOOLS}:
|
||||
return
|
||||
|
||||
if last.get("role") != "tool" and num_turns > MAX_NUM_TURNS:
|
||||
return
|
||||
|
||||
if last.get("role") != "tool" and next_turn_should_call_tools:
|
||||
return
|
||||
|
||||
next_turn_should_call_tools = last.get("role") != "tool"
|
||||
247
venv/Lib/site-packages/huggingface_hub/inference/_mcp/cli.py
Normal file
247
venv/Lib/site-packages/huggingface_hub/inference/_mcp/cli.py
Normal file
|
|
@ -0,0 +1,247 @@
|
|||
import asyncio
|
||||
import os
|
||||
import signal
|
||||
import traceback
|
||||
from typing import Optional
|
||||
|
||||
import typer
|
||||
from rich import print
|
||||
|
||||
from ._cli_hacks import _async_prompt, _patch_anyio_open_process
|
||||
from .agent import Agent
|
||||
from .utils import _load_agent_config
|
||||
|
||||
|
||||
app = typer.Typer(
|
||||
rich_markup_mode="rich",
|
||||
help="A squad of lightweight composable AI applications built on Hugging Face's Inference Client and MCP stack.",
|
||||
)
|
||||
|
||||
run_cli = typer.Typer(
|
||||
name="run",
|
||||
help="Run the Agent in the CLI",
|
||||
invoke_without_command=True,
|
||||
)
|
||||
app.add_typer(run_cli, name="run")
|
||||
|
||||
|
||||
async def run_agent(
|
||||
agent_path: Optional[str],
|
||||
) -> None:
|
||||
"""
|
||||
Tiny Agent loop.
|
||||
|
||||
Args:
|
||||
agent_path (`str`, *optional*):
|
||||
Path to a local folder containing an `agent.json` and optionally a custom `PROMPT.md` file or a built-in agent stored in a Hugging Face dataset.
|
||||
|
||||
"""
|
||||
_patch_anyio_open_process() # Hacky way to prevent stdio connections to be stopped by Ctrl+C
|
||||
|
||||
config, prompt = _load_agent_config(agent_path)
|
||||
|
||||
inputs = config.get("inputs", [])
|
||||
servers = config.get("servers", [])
|
||||
|
||||
abort_event = asyncio.Event()
|
||||
exit_event = asyncio.Event()
|
||||
first_sigint = True
|
||||
|
||||
loop = asyncio.get_running_loop()
|
||||
original_sigint_handler = signal.getsignal(signal.SIGINT)
|
||||
|
||||
def _sigint_handler() -> None:
|
||||
nonlocal first_sigint
|
||||
if first_sigint:
|
||||
first_sigint = False
|
||||
abort_event.set()
|
||||
print("\n[red]Interrupted. Press Ctrl+C again to quit.[/red]", flush=True)
|
||||
return
|
||||
|
||||
print("\n[red]Exiting...[/red]", flush=True)
|
||||
exit_event.set()
|
||||
|
||||
try:
|
||||
sigint_registered_in_loop = False
|
||||
try:
|
||||
loop.add_signal_handler(signal.SIGINT, _sigint_handler)
|
||||
sigint_registered_in_loop = True
|
||||
except (AttributeError, NotImplementedError):
|
||||
# Windows (or any loop that doesn't support it) : fall back to sync
|
||||
signal.signal(signal.SIGINT, lambda *_: _sigint_handler())
|
||||
|
||||
# Handle inputs (i.e. env variables injection)
|
||||
resolved_inputs: dict[str, str] = {}
|
||||
|
||||
if len(inputs) > 0:
|
||||
print(
|
||||
"[bold blue]Some initial inputs are required by the agent. "
|
||||
"Please provide a value or leave empty to load from env.[/bold blue]"
|
||||
)
|
||||
for input_item in inputs:
|
||||
input_id = input_item["id"]
|
||||
description = input_item["description"]
|
||||
env_special_value = f"${{input:{input_id}}}"
|
||||
|
||||
# Check if the input is used by any server or as an apiKey
|
||||
input_usages = set()
|
||||
for server in servers:
|
||||
# Check stdio's "env" and http/sse's "headers" mappings
|
||||
env_or_headers = server.get("env", {}) if server["type"] == "stdio" else server.get("headers", {})
|
||||
for key, value in env_or_headers.items():
|
||||
if env_special_value in value:
|
||||
input_usages.add(key)
|
||||
|
||||
raw_api_key = config.get("apiKey")
|
||||
if isinstance(raw_api_key, str) and env_special_value in raw_api_key:
|
||||
input_usages.add("apiKey")
|
||||
|
||||
if not input_usages:
|
||||
print(
|
||||
f"[yellow]Input '{input_id}' defined in config but not used by any server or as an API key."
|
||||
" Skipping.[/yellow]"
|
||||
)
|
||||
continue
|
||||
|
||||
# Prompt user for input
|
||||
env_variable_key = input_id.replace("-", "_").upper()
|
||||
print(
|
||||
f"[blue] • {input_id}[/blue]: {description}. (default: load from {env_variable_key}).",
|
||||
end=" ",
|
||||
)
|
||||
user_input = (await _async_prompt(exit_event=exit_event)).strip()
|
||||
if exit_event.is_set():
|
||||
return
|
||||
|
||||
# Fallback to environment variable when user left blank
|
||||
final_value = user_input
|
||||
if not final_value:
|
||||
final_value = os.getenv(env_variable_key, "")
|
||||
if final_value:
|
||||
print(f"[green]Value successfully loaded from '{env_variable_key}'[/green]")
|
||||
else:
|
||||
print(
|
||||
f"[yellow]No value found for '{env_variable_key}' in environment variables. Continuing.[/yellow]"
|
||||
)
|
||||
resolved_inputs[input_id] = final_value
|
||||
|
||||
# Inject resolved value (can be empty) into stdio's env or http/sse's headers
|
||||
for server in servers:
|
||||
env_or_headers = server.get("env", {}) if server["type"] == "stdio" else server.get("headers", {})
|
||||
for key, value in env_or_headers.items():
|
||||
if env_special_value in value:
|
||||
env_or_headers[key] = env_or_headers[key].replace(env_special_value, final_value)
|
||||
|
||||
print()
|
||||
|
||||
raw_api_key = config.get("apiKey")
|
||||
if isinstance(raw_api_key, str):
|
||||
substituted_api_key = raw_api_key
|
||||
for input_id, val in resolved_inputs.items():
|
||||
substituted_api_key = substituted_api_key.replace(f"${{input:{input_id}}}", val)
|
||||
config["apiKey"] = substituted_api_key
|
||||
# Main agent loop
|
||||
async with Agent(
|
||||
provider=config.get("provider"), # type: ignore[arg-type]
|
||||
model=config.get("model"),
|
||||
base_url=config.get("endpointUrl"), # type: ignore[arg-type]
|
||||
api_key=config.get("apiKey"),
|
||||
servers=servers, # type: ignore[arg-type]
|
||||
prompt=prompt,
|
||||
) as agent:
|
||||
await agent.load_tools()
|
||||
print(f"[bold blue]Agent loaded with {len(agent.available_tools)} tools:[/bold blue]")
|
||||
for t in agent.available_tools:
|
||||
print(f"[blue] • {t.function.name}[/blue]")
|
||||
|
||||
while True:
|
||||
abort_event.clear()
|
||||
|
||||
# Check if we should exit
|
||||
if exit_event.is_set():
|
||||
return
|
||||
|
||||
try:
|
||||
user_input = await _async_prompt(exit_event=exit_event)
|
||||
first_sigint = True
|
||||
except EOFError:
|
||||
print("\n[red]EOF received, exiting.[/red]", flush=True)
|
||||
break
|
||||
except KeyboardInterrupt:
|
||||
if not first_sigint and abort_event.is_set():
|
||||
continue
|
||||
else:
|
||||
print("\n[red]Keyboard interrupt during input processing.[/red]", flush=True)
|
||||
break
|
||||
|
||||
try:
|
||||
async for chunk in agent.run(user_input, abort_event=abort_event):
|
||||
if abort_event.is_set() and not first_sigint:
|
||||
break
|
||||
if exit_event.is_set():
|
||||
return
|
||||
|
||||
if hasattr(chunk, "choices"):
|
||||
delta = chunk.choices[0].delta
|
||||
if delta.content:
|
||||
print(delta.content, end="", flush=True)
|
||||
if delta.tool_calls:
|
||||
for call in delta.tool_calls:
|
||||
if call.id:
|
||||
print(f"<Tool {call.id}>", end="")
|
||||
if call.function.name:
|
||||
print(f"{call.function.name}", end=" ")
|
||||
if call.function.arguments:
|
||||
print(f"{call.function.arguments}", end="")
|
||||
else:
|
||||
print(
|
||||
f"\n\n[green]Tool[{chunk.name}] {chunk.tool_call_id}\n{chunk.content}[/green]\n",
|
||||
flush=True,
|
||||
)
|
||||
|
||||
print()
|
||||
|
||||
except Exception as e:
|
||||
tb_str = traceback.format_exc()
|
||||
print(f"\n[bold red]Error during agent run: {e}\n{tb_str}[/bold red]", flush=True)
|
||||
first_sigint = True # Allow graceful interrupt for the next command
|
||||
|
||||
except Exception as e:
|
||||
tb_str = traceback.format_exc()
|
||||
print(f"\n[bold red]An unexpected error occurred: {e}\n{tb_str}[/bold red]", flush=True)
|
||||
raise e
|
||||
|
||||
finally:
|
||||
if sigint_registered_in_loop:
|
||||
try:
|
||||
loop.remove_signal_handler(signal.SIGINT)
|
||||
except (AttributeError, NotImplementedError):
|
||||
pass
|
||||
else:
|
||||
signal.signal(signal.SIGINT, original_sigint_handler)
|
||||
|
||||
|
||||
@run_cli.callback()
|
||||
def run(
|
||||
path: Optional[str] = typer.Argument(
|
||||
None,
|
||||
help=(
|
||||
"Path to a local folder containing an agent.json file or a built-in agent "
|
||||
"stored in the 'tiny-agents/tiny-agents' Hugging Face dataset "
|
||||
"(https://huggingface.co/datasets/tiny-agents/tiny-agents)"
|
||||
),
|
||||
show_default=False,
|
||||
),
|
||||
):
|
||||
try:
|
||||
asyncio.run(run_agent(path))
|
||||
except KeyboardInterrupt:
|
||||
print("\n[red]Application terminated by KeyboardInterrupt.[/red]", flush=True)
|
||||
raise typer.Exit(code=130)
|
||||
except Exception as e:
|
||||
print(f"\n[bold red]An unexpected error occurred: {e}[/bold red]", flush=True)
|
||||
raise e
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
app()
|
||||
|
|
@ -0,0 +1,82 @@
|
|||
from __future__ import annotations
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import List
|
||||
|
||||
from huggingface_hub import ChatCompletionInputTool
|
||||
|
||||
|
||||
FILENAME_CONFIG = "agent.json"
|
||||
FILENAME_PROMPT = "PROMPT.md"
|
||||
|
||||
DEFAULT_AGENT = {
|
||||
"model": "Qwen/Qwen2.5-72B-Instruct",
|
||||
"provider": "nebius",
|
||||
"servers": [
|
||||
{
|
||||
"type": "stdio",
|
||||
"command": "npx",
|
||||
"args": [
|
||||
"-y",
|
||||
"@modelcontextprotocol/server-filesystem",
|
||||
str(Path.home() / ("Desktop" if sys.platform == "darwin" else "")),
|
||||
],
|
||||
},
|
||||
{
|
||||
"type": "stdio",
|
||||
"command": "npx",
|
||||
"args": ["@playwright/mcp@latest"],
|
||||
},
|
||||
],
|
||||
}
|
||||
|
||||
|
||||
DEFAULT_SYSTEM_PROMPT = """
|
||||
You are an agent - please keep going until the user’s query is completely
|
||||
resolved, before ending your turn and yielding back to the user. Only terminate
|
||||
your turn when you are sure that the problem is solved, or if you need more
|
||||
info from the user to solve the problem.
|
||||
If you are not sure about anything pertaining to the user’s request, use your
|
||||
tools to read files and gather the relevant information: do NOT guess or make
|
||||
up an answer.
|
||||
You MUST plan extensively before each function call, and reflect extensively
|
||||
on the outcomes of the previous function calls. DO NOT do this entire process
|
||||
by making function calls only, as this can impair your ability to solve the
|
||||
problem and think insightfully.
|
||||
""".strip()
|
||||
|
||||
MAX_NUM_TURNS = 10
|
||||
|
||||
TASK_COMPLETE_TOOL: ChatCompletionInputTool = ChatCompletionInputTool.parse_obj( # type: ignore[assignment]
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "task_complete",
|
||||
"description": "Call this tool when the task given by the user is complete",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {},
|
||||
},
|
||||
},
|
||||
}
|
||||
)
|
||||
|
||||
ASK_QUESTION_TOOL: ChatCompletionInputTool = ChatCompletionInputTool.parse_obj( # type: ignore[assignment]
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "ask_question",
|
||||
"description": "Ask the user for more info required to solve or clarify their problem.",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {},
|
||||
},
|
||||
},
|
||||
}
|
||||
)
|
||||
|
||||
EXIT_LOOP_TOOLS: List[ChatCompletionInputTool] = [TASK_COMPLETE_TOOL, ASK_QUESTION_TOOL]
|
||||
|
||||
|
||||
DEFAULT_REPO_ID = "tiny-agents/tiny-agents"
|
||||
|
|
@ -0,0 +1,369 @@
|
|||
import json
|
||||
import logging
|
||||
from contextlib import AsyncExitStack
|
||||
from datetime import timedelta
|
||||
from pathlib import Path
|
||||
from typing import TYPE_CHECKING, Any, AsyncIterable, Dict, List, Literal, Optional, Union, overload
|
||||
|
||||
from typing_extensions import NotRequired, TypeAlias, TypedDict, Unpack
|
||||
|
||||
from ...utils._runtime import get_hf_hub_version
|
||||
from .._generated._async_client import AsyncInferenceClient
|
||||
from .._generated.types import (
|
||||
ChatCompletionInputMessage,
|
||||
ChatCompletionInputTool,
|
||||
ChatCompletionStreamOutput,
|
||||
ChatCompletionStreamOutputDeltaToolCall,
|
||||
)
|
||||
from .._providers import PROVIDER_OR_POLICY_T
|
||||
from .utils import format_result
|
||||
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from mcp import ClientSession
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Type alias for tool names
|
||||
ToolName: TypeAlias = str
|
||||
|
||||
ServerType: TypeAlias = Literal["stdio", "sse", "http"]
|
||||
|
||||
|
||||
class StdioServerParameters_T(TypedDict):
|
||||
command: str
|
||||
args: NotRequired[List[str]]
|
||||
env: NotRequired[Dict[str, str]]
|
||||
cwd: NotRequired[Union[str, Path, None]]
|
||||
|
||||
|
||||
class SSEServerParameters_T(TypedDict):
|
||||
url: str
|
||||
headers: NotRequired[Dict[str, Any]]
|
||||
timeout: NotRequired[float]
|
||||
sse_read_timeout: NotRequired[float]
|
||||
|
||||
|
||||
class StreamableHTTPParameters_T(TypedDict):
|
||||
url: str
|
||||
headers: NotRequired[dict[str, Any]]
|
||||
timeout: NotRequired[timedelta]
|
||||
sse_read_timeout: NotRequired[timedelta]
|
||||
terminate_on_close: NotRequired[bool]
|
||||
|
||||
|
||||
class MCPClient:
|
||||
"""
|
||||
Client for connecting to one or more MCP servers and processing chat completions with tools.
|
||||
|
||||
<Tip warning={true}>
|
||||
|
||||
This class is experimental and might be subject to breaking changes in the future without prior notice.
|
||||
|
||||
</Tip>
|
||||
|
||||
Args:
|
||||
model (`str`, `optional`):
|
||||
The model to run inference with. Can be a model id hosted on the Hugging Face Hub, e.g. `meta-llama/Meta-Llama-3-8B-Instruct`
|
||||
or a URL to a deployed Inference Endpoint or other local or remote endpoint.
|
||||
provider (`str`, *optional*):
|
||||
Name of the provider to use for inference. Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
|
||||
If model is a URL or `base_url` is passed, then `provider` is not used.
|
||||
base_url (`str`, *optional*):
|
||||
The base URL to run inference. Defaults to None.
|
||||
api_key (`str`, `optional`):
|
||||
Token to use for authentication. Will default to the locally Hugging Face saved token if not provided. You can also use your own provider API key to interact directly with the provider's service.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
model: Optional[str] = None,
|
||||
provider: Optional[PROVIDER_OR_POLICY_T] = None,
|
||||
base_url: Optional[str] = None,
|
||||
api_key: Optional[str] = None,
|
||||
):
|
||||
# Initialize MCP sessions as a dictionary of ClientSession objects
|
||||
self.sessions: Dict[ToolName, "ClientSession"] = {}
|
||||
self.exit_stack = AsyncExitStack()
|
||||
self.available_tools: List[ChatCompletionInputTool] = []
|
||||
# To be able to send the model in the payload if `base_url` is provided
|
||||
if model is None and base_url is None:
|
||||
raise ValueError("At least one of `model` or `base_url` should be set in `MCPClient`.")
|
||||
self.payload_model = model
|
||||
self.client = AsyncInferenceClient(
|
||||
model=None if base_url is not None else model,
|
||||
provider=provider,
|
||||
api_key=api_key,
|
||||
base_url=base_url,
|
||||
)
|
||||
|
||||
async def __aenter__(self):
|
||||
"""Enter the context manager"""
|
||||
await self.client.__aenter__()
|
||||
await self.exit_stack.__aenter__()
|
||||
return self
|
||||
|
||||
async def __aexit__(self, exc_type, exc_val, exc_tb):
|
||||
"""Exit the context manager"""
|
||||
await self.client.__aexit__(exc_type, exc_val, exc_tb)
|
||||
await self.cleanup()
|
||||
|
||||
async def cleanup(self):
|
||||
"""Clean up resources"""
|
||||
await self.client.close()
|
||||
await self.exit_stack.aclose()
|
||||
|
||||
@overload
|
||||
async def add_mcp_server(self, type: Literal["stdio"], **params: Unpack[StdioServerParameters_T]): ...
|
||||
|
||||
@overload
|
||||
async def add_mcp_server(self, type: Literal["sse"], **params: Unpack[SSEServerParameters_T]): ...
|
||||
|
||||
@overload
|
||||
async def add_mcp_server(self, type: Literal["http"], **params: Unpack[StreamableHTTPParameters_T]): ...
|
||||
|
||||
async def add_mcp_server(self, type: ServerType, **params: Any):
|
||||
"""Connect to an MCP server
|
||||
|
||||
Args:
|
||||
type (`str`):
|
||||
Type of the server to connect to. Can be one of:
|
||||
- "stdio": Standard input/output server (local)
|
||||
- "sse": Server-sent events (SSE) server
|
||||
- "http": StreamableHTTP server
|
||||
**params (`Dict[str, Any]`):
|
||||
Server parameters that can be either:
|
||||
- For stdio servers:
|
||||
- command (str): The command to run the MCP server
|
||||
- args (List[str], optional): Arguments for the command
|
||||
- env (Dict[str, str], optional): Environment variables for the command
|
||||
- cwd (Union[str, Path, None], optional): Working directory for the command
|
||||
- For SSE servers:
|
||||
- url (str): The URL of the SSE server
|
||||
- headers (Dict[str, Any], optional): Headers for the SSE connection
|
||||
- timeout (float, optional): Connection timeout
|
||||
- sse_read_timeout (float, optional): SSE read timeout
|
||||
- For StreamableHTTP servers:
|
||||
- url (str): The URL of the StreamableHTTP server
|
||||
- headers (Dict[str, Any], optional): Headers for the StreamableHTTP connection
|
||||
- timeout (timedelta, optional): Connection timeout
|
||||
- sse_read_timeout (timedelta, optional): SSE read timeout
|
||||
- terminate_on_close (bool, optional): Whether to terminate on close
|
||||
"""
|
||||
from mcp import ClientSession, StdioServerParameters
|
||||
from mcp import types as mcp_types
|
||||
|
||||
# Determine server type and create appropriate parameters
|
||||
if type == "stdio":
|
||||
# Handle stdio server
|
||||
from mcp.client.stdio import stdio_client
|
||||
|
||||
logger.info(f"Connecting to stdio MCP server with command: {params['command']} {params.get('args', [])}")
|
||||
|
||||
client_kwargs = {"command": params["command"]}
|
||||
for key in ["args", "env", "cwd"]:
|
||||
if params.get(key) is not None:
|
||||
client_kwargs[key] = params[key]
|
||||
server_params = StdioServerParameters(**client_kwargs)
|
||||
read, write = await self.exit_stack.enter_async_context(stdio_client(server_params))
|
||||
elif type == "sse":
|
||||
# Handle SSE server
|
||||
from mcp.client.sse import sse_client
|
||||
|
||||
logger.info(f"Connecting to SSE MCP server at: {params['url']}")
|
||||
|
||||
client_kwargs = {"url": params["url"]}
|
||||
for key in ["headers", "timeout", "sse_read_timeout"]:
|
||||
if params.get(key) is not None:
|
||||
client_kwargs[key] = params[key]
|
||||
read, write = await self.exit_stack.enter_async_context(sse_client(**client_kwargs))
|
||||
elif type == "http":
|
||||
# Handle StreamableHTTP server
|
||||
from mcp.client.streamable_http import streamablehttp_client
|
||||
|
||||
logger.info(f"Connecting to StreamableHTTP MCP server at: {params['url']}")
|
||||
|
||||
client_kwargs = {"url": params["url"]}
|
||||
for key in ["headers", "timeout", "sse_read_timeout", "terminate_on_close"]:
|
||||
if params.get(key) is not None:
|
||||
client_kwargs[key] = params[key]
|
||||
read, write, _ = await self.exit_stack.enter_async_context(streamablehttp_client(**client_kwargs))
|
||||
# ^ TODO: should be handle `get_session_id_callback`? (function to retrieve the current session ID)
|
||||
else:
|
||||
raise ValueError(f"Unsupported server type: {type}")
|
||||
|
||||
session = await self.exit_stack.enter_async_context(
|
||||
ClientSession(
|
||||
read_stream=read,
|
||||
write_stream=write,
|
||||
client_info=mcp_types.Implementation(
|
||||
name="huggingface_hub.MCPClient",
|
||||
version=get_hf_hub_version(),
|
||||
),
|
||||
)
|
||||
)
|
||||
|
||||
logger.debug("Initializing session...")
|
||||
await session.initialize()
|
||||
|
||||
# List available tools
|
||||
response = await session.list_tools()
|
||||
logger.debug("Connected to server with tools:", [tool.name for tool in response.tools])
|
||||
|
||||
for tool in response.tools:
|
||||
if tool.name in self.sessions:
|
||||
logger.warning(f"Tool '{tool.name}' already defined by another server. Skipping.")
|
||||
continue
|
||||
|
||||
# Map tool names to their server for later lookup
|
||||
self.sessions[tool.name] = session
|
||||
|
||||
# Add tool to the list of available tools (for use in chat completions)
|
||||
self.available_tools.append(
|
||||
ChatCompletionInputTool.parse_obj_as_instance(
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": tool.name,
|
||||
"description": tool.description,
|
||||
"parameters": tool.inputSchema,
|
||||
},
|
||||
}
|
||||
)
|
||||
)
|
||||
|
||||
async def process_single_turn_with_tools(
|
||||
self,
|
||||
messages: List[Union[Dict, ChatCompletionInputMessage]],
|
||||
exit_loop_tools: Optional[List[ChatCompletionInputTool]] = None,
|
||||
exit_if_first_chunk_no_tool: bool = False,
|
||||
) -> AsyncIterable[Union[ChatCompletionStreamOutput, ChatCompletionInputMessage]]:
|
||||
"""Process a query using `self.model` and available tools, yielding chunks and tool outputs.
|
||||
|
||||
Args:
|
||||
messages (`List[Dict]`):
|
||||
List of message objects representing the conversation history
|
||||
exit_loop_tools (`List[ChatCompletionInputTool]`, *optional*):
|
||||
List of tools that should exit the generator when called
|
||||
exit_if_first_chunk_no_tool (`bool`, *optional*):
|
||||
Exit if no tool is present in the first chunks. Default to False.
|
||||
|
||||
Yields:
|
||||
[`ChatCompletionStreamOutput`] chunks or [`ChatCompletionInputMessage`] objects
|
||||
"""
|
||||
# Prepare tools list based on options
|
||||
tools = self.available_tools
|
||||
if exit_loop_tools is not None:
|
||||
tools = [*exit_loop_tools, *self.available_tools]
|
||||
|
||||
# Create the streaming request
|
||||
response = await self.client.chat.completions.create(
|
||||
model=self.payload_model,
|
||||
messages=messages,
|
||||
tools=tools,
|
||||
tool_choice="auto",
|
||||
stream=True,
|
||||
)
|
||||
|
||||
message: Dict[str, Any] = {"role": "unknown", "content": ""}
|
||||
final_tool_calls: Dict[int, ChatCompletionStreamOutputDeltaToolCall] = {}
|
||||
num_of_chunks = 0
|
||||
|
||||
# Read from stream
|
||||
async for chunk in response:
|
||||
num_of_chunks += 1
|
||||
delta = chunk.choices[0].delta if chunk.choices and len(chunk.choices) > 0 else None
|
||||
if not delta:
|
||||
continue
|
||||
|
||||
# Process message
|
||||
if delta.role:
|
||||
message["role"] = delta.role
|
||||
if delta.content:
|
||||
message["content"] += delta.content
|
||||
|
||||
# Process tool calls
|
||||
if delta.tool_calls:
|
||||
for tool_call in delta.tool_calls:
|
||||
# Aggregate chunks into tool calls
|
||||
if tool_call.index not in final_tool_calls:
|
||||
if (
|
||||
tool_call.function.arguments is None or tool_call.function.arguments == "{}"
|
||||
): # Corner case (depends on provider)
|
||||
tool_call.function.arguments = ""
|
||||
final_tool_calls[tool_call.index] = tool_call
|
||||
|
||||
elif tool_call.function.arguments:
|
||||
final_tool_calls[tool_call.index].function.arguments += tool_call.function.arguments
|
||||
|
||||
# Optionally exit early if no tools in first chunks
|
||||
if exit_if_first_chunk_no_tool and num_of_chunks <= 2 and len(final_tool_calls) == 0:
|
||||
return
|
||||
|
||||
# Yield each chunk to caller
|
||||
yield chunk
|
||||
|
||||
# Add the assistant message with tool calls (if any) to messages
|
||||
if message["content"] or final_tool_calls:
|
||||
# if the role is unknown, set it to assistant
|
||||
if message.get("role") == "unknown":
|
||||
message["role"] = "assistant"
|
||||
# Convert final_tool_calls to the format expected by OpenAI
|
||||
if final_tool_calls:
|
||||
tool_calls_list: List[Dict[str, Any]] = []
|
||||
for tc in final_tool_calls.values():
|
||||
tool_calls_list.append(
|
||||
{
|
||||
"id": tc.id,
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": tc.function.name,
|
||||
"arguments": tc.function.arguments or "{}",
|
||||
},
|
||||
}
|
||||
)
|
||||
message["tool_calls"] = tool_calls_list
|
||||
messages.append(message)
|
||||
|
||||
# Process tool calls one by one
|
||||
for tool_call in final_tool_calls.values():
|
||||
function_name = tool_call.function.name
|
||||
try:
|
||||
function_args = json.loads(tool_call.function.arguments or "{}")
|
||||
except json.JSONDecodeError as err:
|
||||
tool_message = {
|
||||
"role": "tool",
|
||||
"tool_call_id": tool_call.id,
|
||||
"name": function_name,
|
||||
"content": f"Invalid JSON generated by the model: {err}",
|
||||
}
|
||||
tool_message_as_obj = ChatCompletionInputMessage.parse_obj_as_instance(tool_message)
|
||||
messages.append(tool_message_as_obj)
|
||||
yield tool_message_as_obj
|
||||
continue # move to next tool call
|
||||
|
||||
tool_message = {"role": "tool", "tool_call_id": tool_call.id, "content": "", "name": function_name}
|
||||
|
||||
# Check if this is an exit loop tool
|
||||
if exit_loop_tools and function_name in [t.function.name for t in exit_loop_tools]:
|
||||
tool_message_as_obj = ChatCompletionInputMessage.parse_obj_as_instance(tool_message)
|
||||
messages.append(tool_message_as_obj)
|
||||
yield tool_message_as_obj
|
||||
return
|
||||
|
||||
# Execute tool call with the appropriate session
|
||||
session = self.sessions.get(function_name)
|
||||
if session is not None:
|
||||
try:
|
||||
result = await session.call_tool(function_name, function_args)
|
||||
tool_message["content"] = format_result(result)
|
||||
except Exception as err:
|
||||
tool_message["content"] = f"Error: MCP tool call failed with error message: {err}"
|
||||
else:
|
||||
tool_message["content"] = f"Error: No session found for tool: {function_name}"
|
||||
|
||||
# Yield tool message
|
||||
tool_message_as_obj = ChatCompletionInputMessage.parse_obj_as_instance(tool_message)
|
||||
messages.append(tool_message_as_obj)
|
||||
yield tool_message_as_obj
|
||||
|
|
@ -0,0 +1,42 @@
|
|||
from typing import Dict, List, Literal, TypedDict, Union
|
||||
|
||||
from typing_extensions import NotRequired
|
||||
|
||||
|
||||
class InputConfig(TypedDict, total=False):
|
||||
id: str
|
||||
description: str
|
||||
type: str
|
||||
password: bool
|
||||
|
||||
|
||||
class StdioServerConfig(TypedDict):
|
||||
type: Literal["stdio"]
|
||||
command: str
|
||||
args: List[str]
|
||||
env: Dict[str, str]
|
||||
cwd: str
|
||||
|
||||
|
||||
class HTTPServerConfig(TypedDict):
|
||||
type: Literal["http"]
|
||||
url: str
|
||||
headers: Dict[str, str]
|
||||
|
||||
|
||||
class SSEServerConfig(TypedDict):
|
||||
type: Literal["sse"]
|
||||
url: str
|
||||
headers: Dict[str, str]
|
||||
|
||||
|
||||
ServerConfig = Union[StdioServerConfig, HTTPServerConfig, SSEServerConfig]
|
||||
|
||||
|
||||
# AgentConfig root object
|
||||
class AgentConfig(TypedDict):
|
||||
model: str
|
||||
provider: str
|
||||
apiKey: NotRequired[str]
|
||||
inputs: List[InputConfig]
|
||||
servers: List[ServerConfig]
|
||||
124
venv/Lib/site-packages/huggingface_hub/inference/_mcp/utils.py
Normal file
124
venv/Lib/site-packages/huggingface_hub/inference/_mcp/utils.py
Normal file
|
|
@ -0,0 +1,124 @@
|
|||
"""
|
||||
Utility functions for MCPClient and Tiny Agents.
|
||||
|
||||
Formatting utilities taken from the JS SDK: https://github.com/huggingface/huggingface.js/blob/main/packages/mcp-client/src/ResultFormatter.ts.
|
||||
"""
|
||||
|
||||
import json
|
||||
from pathlib import Path
|
||||
from typing import TYPE_CHECKING, List, Optional, Tuple
|
||||
|
||||
from huggingface_hub import snapshot_download
|
||||
from huggingface_hub.errors import EntryNotFoundError
|
||||
|
||||
from .constants import DEFAULT_AGENT, DEFAULT_REPO_ID, FILENAME_CONFIG, FILENAME_PROMPT
|
||||
from .types import AgentConfig
|
||||
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from mcp import types as mcp_types
|
||||
|
||||
|
||||
def format_result(result: "mcp_types.CallToolResult") -> str:
|
||||
"""
|
||||
Formats a mcp.types.CallToolResult content into a human-readable string.
|
||||
|
||||
Args:
|
||||
result (CallToolResult)
|
||||
Object returned by mcp.ClientSession.call_tool.
|
||||
|
||||
Returns:
|
||||
str
|
||||
A formatted string representing the content of the result.
|
||||
"""
|
||||
content = result.content
|
||||
|
||||
if len(content) == 0:
|
||||
return "[No content]"
|
||||
|
||||
formatted_parts: List[str] = []
|
||||
|
||||
for item in content:
|
||||
if item.type == "text":
|
||||
formatted_parts.append(item.text)
|
||||
|
||||
elif item.type == "image":
|
||||
formatted_parts.append(
|
||||
f"[Binary Content: Image {item.mimeType}, {_get_base64_size(item.data)} bytes]\n"
|
||||
f"The task is complete and the content accessible to the User"
|
||||
)
|
||||
|
||||
elif item.type == "audio":
|
||||
formatted_parts.append(
|
||||
f"[Binary Content: Audio {item.mimeType}, {_get_base64_size(item.data)} bytes]\n"
|
||||
f"The task is complete and the content accessible to the User"
|
||||
)
|
||||
|
||||
elif item.type == "resource":
|
||||
resource = item.resource
|
||||
|
||||
if hasattr(resource, "text"):
|
||||
formatted_parts.append(resource.text)
|
||||
|
||||
elif hasattr(resource, "blob"):
|
||||
formatted_parts.append(
|
||||
f"[Binary Content ({resource.uri}): {resource.mimeType}, {_get_base64_size(resource.blob)} bytes]\n"
|
||||
f"The task is complete and the content accessible to the User"
|
||||
)
|
||||
|
||||
return "\n".join(formatted_parts)
|
||||
|
||||
|
||||
def _get_base64_size(base64_str: str) -> int:
|
||||
"""Estimate the byte size of a base64-encoded string."""
|
||||
# Remove any prefix like "data:image/png;base64,"
|
||||
if "," in base64_str:
|
||||
base64_str = base64_str.split(",")[1]
|
||||
|
||||
padding = 0
|
||||
if base64_str.endswith("=="):
|
||||
padding = 2
|
||||
elif base64_str.endswith("="):
|
||||
padding = 1
|
||||
|
||||
return (len(base64_str) * 3) // 4 - padding
|
||||
|
||||
|
||||
def _load_agent_config(agent_path: Optional[str]) -> Tuple[AgentConfig, Optional[str]]:
|
||||
"""Load server config and prompt."""
|
||||
|
||||
def _read_dir(directory: Path) -> Tuple[AgentConfig, Optional[str]]:
|
||||
cfg_file = directory / FILENAME_CONFIG
|
||||
if not cfg_file.exists():
|
||||
raise FileNotFoundError(f" Config file not found in {directory}! Please make sure it exists locally")
|
||||
|
||||
config: AgentConfig = json.loads(cfg_file.read_text(encoding="utf-8"))
|
||||
prompt_file = directory / FILENAME_PROMPT
|
||||
prompt: Optional[str] = prompt_file.read_text(encoding="utf-8") if prompt_file.exists() else None
|
||||
return config, prompt
|
||||
|
||||
if agent_path is None:
|
||||
return DEFAULT_AGENT, None # type: ignore[return-value]
|
||||
|
||||
path = Path(agent_path).expanduser()
|
||||
|
||||
if path.is_file():
|
||||
return json.loads(path.read_text(encoding="utf-8")), None
|
||||
|
||||
if path.is_dir():
|
||||
return _read_dir(path)
|
||||
|
||||
# fetch from the Hub
|
||||
try:
|
||||
repo_dir = Path(
|
||||
snapshot_download(
|
||||
repo_id=DEFAULT_REPO_ID,
|
||||
allow_patterns=f"{agent_path}/*",
|
||||
repo_type="dataset",
|
||||
)
|
||||
)
|
||||
return _read_dir(repo_dir / agent_path)
|
||||
except Exception as err:
|
||||
raise EntryNotFoundError(
|
||||
f" Agent {agent_path} not found in tiny-agents/tiny-agents! Please make sure it exists in https://huggingface.co/datasets/tiny-agents/tiny-agents."
|
||||
) from err
|
||||
Loading…
Add table
Add a link
Reference in a new issue