Integrating MCP
Plug-and-play context with the Model Context Protocol
Large-language-model agents can only be as helpful as the context they can see. The Model Context Protocol (MCP) is an open specification that acts like a universal "USB-C port" for AI—standardising how agents connect to files, APIs and tools. Read the full overview in the MCP introduction.
Why MCP?
MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:
- A growing list of pre-built integrations that your LLM can directly plug into
- The flexibility to switch between LLM providers and vendors
- Best practices for securing your data within your infrastructure
ThinAgents + MCP
from thinagents import Agent
agent = Agent(
name="MCP Agent",
model="openai/gpt-4o-mini",
mcp_servers=[
{ # Remote services over SSE
"transport": "sse",
"url": "http://localhost:8100/sse",
},
{ # Local filesystem access over stdio
"transport": "stdio",
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/Users/prabhukirankonda/Documents/projects/mcp_test_2_weather",
],
},
],
)Run It Asynchronously
async for chunk in agent.astream(
"what is the weather in seattle and list the files in the directory that you have access to",
):
print(chunk.content, end="", flush=True)
# Or get the full response at once (non-streaming but still async)
response = await agent.arun(
"what is the weather in seattle and list the files in the directory that you have access to",
)
print(response.content)Async by Design
Agents that use MCP must be invoked with asynchronous APIs. Stream chunks with async for over agent.astream(...) or await a single buffered reply via await agent.arun(...).
Supported Transports in ThinAgents
| Transport | Best For | How to Configure |
|---|---|---|
| SSE | HTTP servers that push events (public APIs, remote micro-services) | { "transport": "sse", "url": "…" } |
| stdio | Local binaries or Node packages you want to spawn (filesystem, CLI tools) | { "transport": "stdio", "command": "…", "args": [ … ] } |