MCP Chain

A composable middleware framework for building MCP server chains, inspired by Ruby Rack. MCP Chain lets you create transparent proxies that sit between MCP clients and servers, transforming requests and responses using Python functions.

MCP Chain solves the problem of adding cross-cutting concerns (authentication, logging, request transformation) to existing MCP servers without modifying them. It uses a transparent proxy pattern where each middleware layer appears as a standard MCP server to clients while forwarding requests to downstream servers. Middleware can also orchestrate multiple MCP calls behind the scenes using AI, transforming granular APIs into intelligent MCPs that perform complex multi-step tasks.

Install and run withuvx- no setup required:

# cli_server.py from mcp_chain import mcp_chain, CLIMCPServer cli_server = CLIMCPServer( name="dev-tools", commands=["git", "ls", "grep"], descriptions={ "git": "Git version control operations", "ls": "List directory contents", "grep": "Search text patterns" } ) # Auto-detected by CLI chain = mcp_chain().then(cli_server)

{ "mcpServers": { "dev-tools": { "command": "uvx", "args": ["mcp-chain", "cli_server.py"] } } }

from mcp_chain import mcp_chain, ExternalMCPServer, serve def require_auth(next_server, request_dict): if not request_dict.get("auth_token"): return {"error": "Authentication required", "code": 401} return next_server.handle_request(request_dict) chain = (mcp_chain() .then(None, require_auth) .then(ExternalMCPServer("postgres", "postgres-mcp"))) serve(chain, name="Authenticated Postgres")

from mcp_chain import mcp_chain, CLIMCPServer def add_company_context(next_server, metadata_dict): metadata = next_server.get_metadata() for tool in metadata.get("tools", []): tool["description"] = f"ACME Corp: {tool.get('description', '')}" return metadata def add_headers(next_server, request_dict): request_dict["headers"] = {"X-Company": "ACME"} response = next_server.handle_request(request_dict) response["processed_by"] = "acme-proxy" return response cli_server = CLIMCPServer(name="tools", commands=["git", "docker"]) chain = (mcp_chain() .then(add_company_context, add_headers) .then(cli_server))

Stack authentication, logging, and transformation:

import logging from mcp_chain import mcp_chain, ExternalMCPServer, serve logging.basicConfig(level=logging.INFO) logger = logging.getLogger("mcp-chain") def auth_middleware(next_server, request_dict): if not request_dict.get("auth_token"): return {"error": "Authentication required", "code": 401} return next_server.handle_request(request_dict) def logging_middleware(next_server, request_dict): logger.info(f"Request: {request_dict.get('method')}") response = next_server.handle_request(request_dict) logger.info(f"Response: {response.get('result', 'error')}") return response def context_middleware(next_server, metadata_dict): metadata = next_server.get_metadata() for tool in metadata.get("tools", []): tool["description"] = f"Enterprise: {tool.get('description', '')}" return metadata chain = (mcp_chain() .then(context_middleware, auth_middleware) .then(None, logging_middleware) .then(ExternalMCPServer("postgres", "postgres-mcp"))) serve(chain, name="Enterprise Postgres")

from mcp_chain import mcp_chain, CLIMCPServer, serve def rate_limit_middleware(next_server, request_dict): # Add rate limiting logic return next_server.handle_request(request_dict) cli_server = CLIMCPServer(name="secure-tools", commands=["git"]) chain = mcp_chain().then(None, rate_limit_middleware).then(cli_server) serve(chain, name="Rate Limited Tools", port=8000)

MCP Chain uses a functional middleware pattern where each layer transforms requests/responses and forwards to the next layer:

graph TD A["MCP Client"] --> B["FastMCP"] B --> C["mcp_chain()"] C --> D1["middleware_1"] D1 --> D2["middleware_2"] D2 --> E["downstream_server"] E -- "response" --> D2 D2 -- "response" --> D1 D1 -- "response" --> C C -- "response" --> B B -- "response" --> A

- Receives requests from the previous layer (or client)
- Transforms the request/metadata using Python dictionaries
- Forwards to the next layer (or downstream server)
- Receives the response back
- Transforms the response as needed
- Returns to the previous layer (or client)

- Transparent Proxy: Each middleware appears as a standard MCP server to clients
- Dict-Based Processing: Internal processing uses Python dicts, not JSON strings
- Composable: Middleware can chain together since each layer is an MCP server
- Zero Overhead: No serialization/deserialization in the middleware chain

Built on the official FastMCP SDK for complete MCP protocol compliance.

from mcp_chain import mcp_chain, serve, CLIMCPServer, ExternalMCPServer # Create a chain chain = mcp_chain() # Add middleware layers chain = chain.then(metadata_transformer, request_transformer) chain = chain.then(downstream_server) # Start server serve(chain, name="My Server", port=8000)

# Metadata transformer (transforms server capabilities) def metadata_transformer(next_server, metadata_dict): metadata = next_server.get_metadata() # Transform metadata dict and return return metadata # Request transformer (transforms requests/responses) def request_transformer(next_server, request_dict): # Transform request dict response = next_server.handle_request(request_dict) # Transform response dict and return return response # Add to chain chain = mcp_chain().then(metadata_transformer, request_transformer)

# CLI server - exposes command-line tools as MCP tools cli_server = CLIMCPServer( name="my-tools", commands=["git", "docker", "npm"], descriptions={ "git": "Git operations", "docker": "Container management", "npm": "Package management" } ) # External server proxy external_server = ExternalMCPServer("server-name", "command-to-run")

The CLI automatically detects chain variables in your Python files:

# Any of these variable names work: chain = mcp_chain().then(...) my_chain = mcp_chain().then(...) server_chain = mcp_chain().then(...) proxy = mcp_chain().then(...)

This project was developed primarily using AI assistants and is designed for AI-assisted development workflows. The codebase is structured to be easily understood and modified by AI tools. Theai/folder contains context documents and design notes specifically for AI assistants working on this repository.

# Development install git clone https://github.com/ronie-uliana/mcp-chain cd mcp-chain uv sync

# Fast unit tests uv run pytest tests/ -m "not integration" -v # Integration tests (with timeout protection) timeout 30 uv run pytest tests/ -m integration -v # All tests timeout 45 uv run pytest tests/ -v # Local CI pipeline ./scripts/test-ci.sh

Releases are automatically published to PyPI via GitHub Actions on new releases.

# Recommended: Run with uvx (no installation) uvx mcp-chain my_chain.py # Install from PyPI pip install mcp-chain # Run installed version python -m mcp_chain my_chain.py # or mcp-chain my_chain.py

This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.

The MCP server for Bitrix24 provides AI assistants with structured access to the Bitrix24 API. It delivers up-to-date method descriptions, parameters, and valid values, allowing assistants to work with precise data instead of guesswork. This reduces code errors and accelerates Bitrix24 integration development.

One remote MCP server for 500+ production APIs — Stripe, HubSpot, Postgres, Gmail, and more. OAuth and API key auth, credential management, and a CLI.

Single tool to control all 100+ API integrations, and UI components

Agent-native developer Q&A API with MCP + A2A endpoints for citations, job pickup, and answer submission.

Self-hosted MCP gateway: convert REST/SOAP/GraphQL/SQL APIs into MCP tools with 29 pre-built adapters, OAuth2, RBAC and audit log.

A universal bridge to convert any web API into an MCP server, supporting multiple transport types.

Dynamically creates MCP servers from web API configurations, integrating any REST API, GraphQL endpoint, or web service into MCP-compatible tools.

Hosted MCP server and coordination layer for AI coding agents — live API contracts, database schema, frontend/backend mismatch detection, and shared handoff tickets for Claude Code, Cursor, Codex, and Lovable.

An MCP server that dynamically loads tools from an external JSON file configured via an environment variable.

A lightweight server exposing Axone's capabilities through the Model-Context Protocol.