OmniMCP Server

OmniMCP is a semantic gateway that unifies multiple MCP servers behind a single entry point. You discover and execute tools across your MCP ecosystem without loading every tool definition into context, keeping prompts lean while enabling cross-server orchestration and progressive loading of tool schemas.

How to use

You interact with a single gateway tool named semantic_router to search for relevant tools, inspect server capabilities, and execute actions across your MCP servers. Start by indexing your MCP servers, then connect to them through OmniMCP using either the standard HTTP transport or the local stdio transport. Use semantic search to find the right tool set by intent, not by exact name, and let the system load only the necessary schemas for execution. For long-running tasks, opt into background execution so you can continue the conversation while results are produced.

Key workflows you can perform include semantic discovery across all indexed tools, obtaining detailed tool schemas just-in-time before execution, starting or stopping servers, and retrieving offloaded content for large results. Use the HTTP transport for remote access and multi-client sharing, or the stdio transport for tight integration with local MCP clients.

How to install

Prerequisites: you need a recent Python and/or Node environment depending on how you run MCP servers, plus a tool for managing process transport (uvx in these examples). Ensure you have a working internet connection to install dependencies and pull container images if you choose Docker.

# Install the CLI tooling (example prerequisites)
# If you use uvx as shown in examples:
# Install uvx if needed (follow your platform's instructions)

# Install OmniMCP package (via Python/Node tooling shown in the docs)
# Example placeholder commands; adapt to your environment
uv pip install omnimcp
# or
uv add omnimcp

Configuration and usage notes

You define your MCP servers in a configuration file that lists each server and how to connect to it. OmniMCP supports two transport modes: stdio (local subprocess) and HTTP (remote server). The configuration can mix both transport types in a single file.

Security and environment

Protect your API keys and tokens. Store sensitive values in environment variables or a secure file and reference them through your config. When using HTTP transport, you may pass authorization headers as part of the server configuration. When using stdio transport, include necessary environment variables in the server config and in the client application.

Notes on content and execution

Large results are managed through content offloading and chunking, with a reference returned for retrieval. You can request content via a reference and chunk index if needed. Background tasks provide a non-blocking workflow for long-running tools.

Available tools

semantic_router

Universal gateway tool that discovers and coordinates all MCP tools across servers. It exposes operations like search_tools, get_server_info, and execute_tool as a single entry point.

search_tools

Semantic search across all indexed tools to find relevant capabilities based on user intent.

get_server_info

Return server capabilities and limitations for a given MCP server.

list_server_tools

Browse tools available on a specific server.

get_tool_details

Retrieve full JSON schema and parameters for a tool before execution.

manage_server

Start or shutdown server instances within the MCP ecosystem.

list_running_servers

Show active servers currently indexed and running.

execute_tool

Run a tool with specified parameters, with optional background execution.

poll_task_result

Check the status and results of a background task.

get_content

Retrieve offloaded content by reference (chunked text, images, or audio).