Meta MCP Server

Meta MCP Server (also known as MCP Deck) is an MCP server that acts as an intelligent router and multiplexer for child Model Context Protocol servers. It connects upstream clients like Claude Desktop and Claude Code to multiple downstream MCP servers through a single connection. Designed for developers and AI engineers managing dozens of individual tools, the server eliminates context-window clutter by indexing tool descriptions and exposing a dynamic tool discovery mechanism. Instead of loading every downstream schema into the client prompt simultaneously, it publishes namespaced tools and a specialized search function. The server utilizes a local Qdrant vector database alongside LM Studio or sentence-transformers to calculate embeddings for semantic tool selection. It also provides an optional RAG pipeline that indexes child-server documentation and an LLM-based selector. When needed, users can run a standalone Gradio dashboard to inspect child server health, debug tool retrieval, and evaluate selection metrics outside standard client supervisors.

Category: AI & LLM Tooling

Tags: embeddings, lm-studio, qdrant, routing, vector-database

Visit Meta MCP Server

How to install and configure Meta MCP Server

1. Ensure Python 3.11+, uv, and Docker or Apple Container (for Qdrant) are installed. 2. Open your client configuration file, such as Claude Desktop's claude_desktop_config.json or Claude Code's .mcp.json. 3. Add mcpdeck to your mcpServers object using the following configuration: json { "mcpServers": { "mcpdeck": { "command": "uvx", "args": [ "--from", "git+https://github.com/anirudhlath/mcpdeck", "mcpdeck", "serve", "--mcp-servers-json", "/absolute/path/to/mcp-servers.json" ] } } } 4. Specify the path to your downstream tool configuration using --mcp-servers-json. 5. Restart Claude Desktop or Claude Code to load the routed MCP tools.

What you can do with Meta MCP Server

• Querying the find_tools meta-tool using natural language to retrieve relevant tools dynamically instead of saturating prompt context windows. * Multiplexing dozens of downstream MCP servers into Claude Desktop through a single stdio connection to prevent client configuration overload. * Running vector search against child-server capabilities using a local Qdrant database and LM Studio embedding pipelines. * Monitoring child server lifecycles, routing strategies, and tool invocation metrics using the standalone Gradio web interface via mcpdeck start.

Key facts

• https://github.com/anirudhlath/meta-mcp
• AI & LLM Tooling, AI Memory & Context, Databases & Data Stores
• embeddings, lm-studio, qdrant, routing, vector-database

Part of MCP Servers

What is Meta MCP Server?

Meta MCP Server, also known as MCP Deck, is an intelligent routing server for the Model Context Protocol. It sits between an MCP client and downstream child servers, using semantic vector search and local embeddings to discover and invoke tools dynamically instead of loading every child tool directly into the client prompt.

Which MCP clients work with Meta MCP Server?

It functions over standard input and output (stdio), making it compatible with Claude Desktop, Claude Code, and any other MCP client that manages child processes using JSON-RPC standard IO pipes.

What can Meta MCP Server do?

It aggregates multiple downstream MCP tools under unique namespaced identifiers, runs a local Qdrant vector database to perform semantic tool searches, offers fallback embeddings via sentence-transformers or LM Studio, and provides a web dashboard for debugging tool retrieval strategies.

How do child tools get named in Meta MCP Server?

Downstream child tools are mapped using the format server_name double underscore tool_name. Dots and disallowed characters are replaced with hyphens, and the full identifier is truncated to the standard MCP limit of 64 characters.