MCPager 📟

A lightweight synchronous client for Model Context Protocol (MCP) servers

MCPager is a small, fast, and predictable Python client for interacting with Model Context Protocol (MCP) servers.
It is designed for independent AI developers who want to connect tools, agents, and LLMs to MCP backends without asyncio, event loops, or framework lock-in.

MCPager acts like a pager: it sends a request, waits for a response, and delivers it reliably — whether the MCP server is running over stdio, HTTP, or streaming (SSE).

---

Why MCPager?

Most MCP client libraries today are:

• deeply async,
• tightly coupled to one agentic framework, if any (e.g. LangChain),
• or built around complex lifecycle management.

MCPager takes a different approach:

Simple. Synchronous. Composable.

• No asyncio
• No background event loops
• Provides abstract adapter for an agentic implementation of your choice (although a concrete LangChain adapter also available out of the box)
• Designed for tool-calling agents

---

Features

• 🧠 Session-based MCP protocol
• 🔌 Multiple transports
  • stdio (local processes)
  • http (FastMCP, remote servers)
  • text/event-stream (SSE / streaming MCP)
• 🧰 Tool discovery & invocation
• 🧩 LangGraph & LangChain integration
• 🧵 Thread-safe synchronous request handling
• 🔍 Built-in JSON-RPC and session ID management

---

Installation

pip install mcpager

Quick Example

1. Start an MCP server

For example, using FastMCP:

fastmcp run examples/echo_server.py --transport http

2. Connect using MCPager

from mcpager.client import StdioTransport, HttpTransport, Session, MCPClient
from mcpager.adapters import LangGraphBackend

transport = HttpTransport('http://localhost:8000/mcp')
# or you can use StdioTransport for local testing like this:
# transport = StdioTransport(["fastmcp", "run", "echo_server.py"])
session = Session(transport)
backend = LangGraphBackend()
client = MCPClient(backend, session)

client.initialize()

tools = client.list_tools()
print(tools)

result = client.call_tool("echo", text="Hello from MCPager!")
print(result)

Using MCPager with LangGraph

MCPager can expose MCP tools directly to LangGraph agents.

from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini")

agent = create_react_agent(llm, tools)

response = agent.invoke(
   {
         'messages': [
            {
               'role': 'user',
               'content': "Please echo the text 'Hello from MCPClient!'"
            }
         ]
   }
)

print("Agent response:", response)

Your MCP tools are now callable by the agent as native LangGraph tools.

Architecture

MCPager is built around three simple layers:

MCPClient
   ↓
BackendAdapter (LangGraphBackend, etc.)
   ↓
Session (JSON-RPC + session handling)
   ↓
Transport (stdio, http, SSE)

This makes MCPager:

• easy to test

• easy to extend

• safe to embed in agents, workers, or pipelines

Transports

Stdio (local MCP servers)

StdioTransport(["python", "my_server.py"])

HTTP (FastMCP or remote MCP)

HttpTransport("http://localhost:3000")

Supports:

• normal JSON responses

• text/event-stream (SSE streaming)

Documentation

Full documentation and guides are available here:

👉 https://uzoo.ai/agents/mcp

This includes:

• MCP protocol overview

• Tool schemas

• Server & client examples

• Agent integration patterns

Who is MCPager for?

MCPager is for:

• Independent AI developers

• Agent builders

• Tool & plugin authors

• Anyone building MCP-based systems without wanting async complexity

If you want reliable tool calls from LLM agents to real backends, MCPager is built for you.

Project Philosophy

MCPager follows a few strict principles:

• Synchronous by default

• Explicit session handling

• No hidden magic

• Transport-agnostic

• Agent-friendly

Think of it as:

“The curl of MCP.”

License

Apache 2.0

Contributions

PRs, issues, and ideas are welcome. MCPager is meant to grow with the MCP ecosystem. 📟

Below is our current backlog:

---

• Task: support sampling
• Task: support elicitation
• 30.12.2025 MCPager client released!