tumiki-proxy
Transparent MCP logging proxy with multi-transport support (stdio, HTTP/StreamableHTTP, HTTP/SSE)
claude mcp add --transport stdio rayven122-tumiki-proxy ./tumiki-proxy --http https://mcp.context7.com/mcp \ --env TUMIKI_LOG_FILE="/tmp/mcp-context7.log" \ --env CONTEXT7_API_KEY="your-api-key"
How to use
tumiki-proxy is a transparent MCP traffic logging proxy for Claude Code and backend MCP servers. It records all MCP traffic to local NDJSON-formatted logs, supporting multiple transports (stdio, HTTP/StreamableHTTP, and HTTP/SSE) and providing optional authentication for HTTP-based servers. The proxy acts as a zero-impact wrapper that does not modify MCP protocol messages, enabling efficient asynchronous buffering and batched writes for debugging and analysis. You can run separate configured instances for local stdio usage (e.g., filesystem logging) or remote MCP servers (e.g., Context7) via HTTP/StreamableHTTP or SSE transports. To use, configure mcpServers in a .mcp.json file or pass equivalent command-line options, set the required environment variables (such as the log file path and API keys), and start the proxy to begin logging traffic between Claude Code and your MCP backend.
How to install
Prerequisites:
- A supported runtime (Bun + binary or Node.js environment) or a prebuilt binary for your platform.
- Basic shell access to install/run commands.
Installation steps (recommended paths):
-
Install via Homebrew (macOS/Linux) if you prefer a package manager:
- brew tap rayven122/tumiki-proxy https://github.com/rayven122/tumiki-proxy
- brew install tumiki-proxy
- brew update && brew upgrade tumiki-proxy
- See the Homebrew install guide for details.
-
Download a prebuilt binary from releases (example for macOS/Linux/Windows):
- curl -L -o tumiki-proxy https://github.com/rayven122/tumiki-proxy/releases/latest/download/tumiki-proxy-linux-x64
- chmod +x tumiki-proxy
-
or appropriate binary for your platform
-
Build from source (recommended for development):
- git clone https://github.com/rayven122/tumiki-proxy.git
- cd tumiki-proxy
- Bun-based build (recommended):
bun install
bun run build
bun run build:binary
Generates a standalone tumiki-proxy binary
- Node.js-based build (alternative):
npm install
npm run build
Run via Node.js: node dist/index.js [args...]
-
Run the proxy with a sample configuration:
- Create a .mcp.json in the project root or run inline with command-line options as shown in the README.
- Example (stdio mode for local MCP server): export TUMIKI_LOG_FILE="./mcp-filesystem.log" ./tumiki-proxy npx -y @modelcontextprotocol/server-filesystem /path/to/dir
Prerequisites recap:
- A runtime compatible with the chosen distribution (Bun binary or Node.js 18+).
- Access to a shell to run the commands above.
- Adequate permissions for creating log files as specified by TUMIKI_LOG_FILE.
Additional notes
Tips and common issues:
- Ensure TUMIKI_LOG_FILE is set and writable; otherwise logs will not be created.
- For HTTP-based transports, supply proper API keys via CONTEXT7_API_KEY, MCP_API_KEY, or API_KEY depending on the server, and verify network access to the MCP backend.
- If StreamableHTTP fails, the proxy will automatically fall back to SSE transport; check logs for messages like 'StreamableHTTP connection failed, falling back to SSE transport'.
- If logs are not appearing, confirm the correct mcpServers configuration in .mcp.json (or the equivalent runtime flags) and verify that the command path to the tumiki-proxy binary is correct.
- The logs use NDJSON; you can grep or parse lines to analyze requests, responses, and transport events.
- When using the binary, provide a relative or absolute path to the executable in the command field of the config to ensure correct startup in different environments.
Related MCP Servers
augments
Comprehensive MCP server providing real-time framework documentation access for Claude Code with intelligent caching, multi-source integration, and context-aware assistance.
docmole
Dig through any documentation with AI - MCP server for Claude, Cursor, and other AI assistants
mcp-graphql-forge
MCP that can proxy any GraphQL API and expose graphql operations as mcp tools.
cdp-tools
MCP server that connects AI assistants to Chrome DevTools Protocol for runtime debugging - set breakpoints, inspect variables, monitor network traffic, and automate browser interactions
mcp-local-llm
MCP server for delegating mechanical tasks to local LLMs via Ollama. Claude does the thinking, your local model does the grunt work.
cogmemai
28 MCP tools that give Ai coding assistants persistent memory across sessions. Works with Claude Code, Cursor, Windsurf, Cline, and Continue. Cloud memory — perfect for teams.