Multi MCP

What religa/multi-mcp MCP server does

religa/multi-mcp MCP server exposes multi-model analysis workflows through the Model Context Protocol. Its main use is coordinating AI-assisted development tasks rather than serving as a general-purpose model API. The available workflows cover interactive chat, code review, comparison of alternatives, and multi-agent debate.

Code review can examine code quality, maintainability, performance, and OWASP Top 10 security concerns. Compare sends a question to multiple configured models to gather separate perspectives. Debate adds an independent-answer and critique process intended to produce a more considered consensus. Chat provides repository-aware development assistance when used from a supported coding client.

The server supports API-backed models from OpenAI, Anthropic, Google, and OpenRouter. It can also run configured command-line models, allowing API and CLI model entries to participate in the same compare or debate workflows.

How it works

An MCP client connects to the server over a local process. The README documents Claude Code and OpenCode integrations. After installation, a user invokes the tools through natural-language requests such as asking for a code review of a module or comparing approaches for a React application.

Model requests are handled asynchronously, so several models can be queried in parallel. The documented performance model is based on the slowest participating model rather than the sum of all model response times. Conversation threading keeps context available across multi-step reviews.

Models are selected by configured names or aliases. Built-in aliases include examples such as mini, sonnet, and gemini, while user-defined models and aliases can be added through YAML configuration. CLI model definitions specify a command, arguments, and parser so they can be invoked as subprocesses alongside API models.

Setup and configuration

religa/multi-mcp MCP server requires Python 3.11 or newer and an API key for at least one supported provider. Source installation uses uv sync, while the repository's make install workflow can install dependencies, create an environment file, update Claude Code or OpenCode configuration, and test the installation.

For manual setup, copy .env.example to .env and add provider credentials. Environment variables take precedence over a project .env; a user-level file at ~/.multi_mcp/.env is used as a fallback. The configuration supports default model and model-list settings, plus optional Azure OpenAI and AWS Bedrock credentials.

The packaged model definitions can be overridden with ~/.multi_mcp/config.yaml. User configuration can add models, define aliases, configure CLI commands, or override settings on existing models. Claude Code can also be configured with an allowlist for the MCP tools and longer timeout values.

Tools and capabilities

The documented MCP tools are:

• chat for interactive development assistance.
• codereview for code-quality, performance, maintainability, and OWASP-oriented analysis.
• compare for parallel responses from multiple models.
• debate for independent responses followed by critique.
• models for listing available models and aliases.

Limitations and notes

The server does not provide its own AI models; it depends on configured provider APIs or locally available CLI model commands. At least one provider credential is needed for the documented API-based setup. CLI models may require separate command-line tools and their own configuration.

The supplied setup examples use a local clone path and a Python module entry point, so paths must be replaced with the actual installation location. The README documents Claude Code and OpenCode, not Claude Desktop, Cursor, Windsurf, or Cline. Model names and aliases are configuration-dependent and may change as user overrides are applied.