Source code for camel.agents.mcp_agent
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
import json
import re
from typing import Optional
from camel.agents import ChatAgent
from camel.logger import get_logger
from camel.messages import BaseMessage
from camel.models import BaseModelBackend
from camel.prompts import TextPrompt
from camel.responses import ChatAgentResponse
from camel.toolkits import MCPToolkit
from camel.types import RoleType
# AgentOps decorator setting
try:
import os
if os.getenv("AGENTOPS_API_KEY") is not None:
from agentops import track_agent
else:
raise ImportError
except (ImportError, AttributeError):
from camel.utils import track_agent
logger = get_logger(__name__)
SYS_MSG_CONTENT = """
You are a helpful assistant, and you prefer to use tools provided by the user
to solve problems.
Using a tool, you will tell the user `server_idx`, `tool_name` and
`tool_args` formatted in JSON as following:
```json
{
"server_idx": idx,
"tool_name": "tool_name",
"tool_args": {
"arg1": value1,
"arg2": value2,
...
}
}
```
For example:
```json
{
"server_idx": 0,
"tool_name": "multiply",
"tool_args": {"a": 5, "b": 50}
}
```
Otherwise, you should respond to the user directly.
"""
TOOLS_PROMPT = """
## Available Tools:
{tools}
"""
FINAL_RESPONSE_PROMPT = """
The result `{results}` is provided by tools you proposed.
Please answer me according to the result directly.
"""
[docs]
@track_agent(name="MCPAgent")
class MCPAgent(ChatAgent):
def __init__(
self,
config_path: str,
model: Optional[BaseModelBackend] = None,
function_calling_available: bool = False,
) -> None:
r"""A class for the MCP agent that assists using MCP tools.
Args:
config_path (str): Path to the MCP configuration file.
model (Optional[BaseModelBackend]): Model backend for the agent.
(default: :obj:`None`)
function_calling_available (bool): Flag indicating whether the
model is equipped with the function calling ability.
(default: :obj:`False`)
"""
if function_calling_available:
sys_msg_content = "You are a helpful assistant, and you prefer "
"to use tools provided by the user to solve problems."
else:
sys_msg_content = SYS_MSG_CONTENT
system_message = BaseMessage(
role_name="MCPRouter",
role_type=RoleType.ASSISTANT,
meta_dict=None,
content=sys_msg_content,
)
super().__init__(system_message, model=model)
self._mcp_toolkit = MCPToolkit(config_path=config_path)
self._function_calling_available = function_calling_available
self._text_tools = None
[docs]
async def connect(self):
r"""Explicitly connect to all MCP servers."""
await self._mcp_toolkit.connect()
[docs]
async def close(self):
r"""Explicitly disconnect from all MCP servers."""
await self._mcp_toolkit.disconnect()
[docs]
def add_mcp_tools(self):
r"""Get the MCP tools and wrap into the models"""
if not self._mcp_toolkit.is_connected():
raise RuntimeError("The MCP server is not connected")
prompt = TextPrompt(TOOLS_PROMPT)
self._text_tools = prompt.format(
tools=self._mcp_toolkit.get_text_tools()
)
if self._function_calling_available:
tools = self._mcp_toolkit.get_tools()
for tool in tools:
self.add_tool(tool)
[docs]
async def run(
self,
prompt: str,
) -> ChatAgentResponse:
r"""Run the agent to interact with the MCP tools.
Args:
prompt (str): The user's input prompt or query to be processed by
the agent.
Returns:
ChatAgentResponse: The agent's response after processing the
prompt and potentially executing MCP tool calls.
Raises:
RuntimeError: If the MCP server is not connected when attempting
to run.
"""
if not self._mcp_toolkit.is_connected():
raise RuntimeError("The MCP server is not connected")
if self._function_calling_available:
response = await self.astep(prompt)
return response
else:
task = f"## Task:\n {prompt}"
response = await self.astep(str(self._text_tools) + task)
content = response.msgs[0].content.lower()
tool_calls = []
while "```json" in content:
json_match = re.search(r'```json', content)
if not json_match:
break
json_start = json_match.span()[1]
end_match = re.search(r'```', content[json_start:])
if not end_match:
break
json_end = end_match.span()[0] + json_start
tool_json = content[json_start:json_end].strip('\n')
try:
tool_calls.append(json.loads(tool_json))
except json.JSONDecodeError:
logger.warning(f"Failed to parse JSON: {tool_json}")
continue
content = content[json_end:]
if not tool_calls:
return response
else:
tools_results = []
for tool_call in tool_calls:
server_idx = tool_call['server_idx']
tool_name = tool_call['tool_name']
tool_args = tool_call['tool_args']
server = self._mcp_toolkit.servers[server_idx]
result = await server.call_tool(tool_name, tool_args)
tools_results.append({tool_name: result.content[0].text})
results = json.dumps(tools_results)
final_prompt = TextPrompt(FINAL_RESPONSE_PROMPT).format(
results=results
)
response = await self.astep(final_prompt)
return response
[docs]
@classmethod
async def create(
cls,
config_path: str,
model: Optional[BaseModelBackend] = None,
function_calling_available: bool = False,
) -> "MCPAgent":
r"""Factory method to create and initialize an MCPAgent.
Args:
config_path (str): Path to the MCP configuration file that contains
server settings and other configuration parameters.
model (Optional[BaseModelBackend]): Model backend for the agent.
If None, the default model will be used. (default: :obj:`None`)
function_calling_available (bool): Flag indicating whether the
model is equipped with function calling ability. This affects
the system message content. (default: :obj:`False`)
Returns:
MCPAgent: A fully initialized MCPAgent instance ready for use.
"""
agent = cls(config_path, model, function_calling_available)
await agent.connect()
agent.add_mcp_tools()
return agent
