Deep Research MCP: One MCP Server for OpenAI, Gemini, Tavily, and Self-Hosted Research Backends

pminervini/deep-research-mcp is a Python package that puts deep research behind the Model Context Protocol. Its FastMCP server in mcp_server.py exposes research as MCP tools, while agent.py handles orchestration and delegates provider-specific work to separate backends. For a builder already working in Claude Code, or another MCP-capable assistant, that means research can be invoked from the assistant session instead of through a separate script or API client. The project is aimed at developers who want provider choice without writing their own MCP layer. Configuration selects the backend, and the same server entry point can be reused across providers. The README lists six options: openai, openai-codex, gemini, tavily, dr-tulu, and gnosis. Providers behind one interface OpenAI via the Responses API with web search and code interpreter, or via Chat Completions for broader compatibility with providers such as Perplexity, Groq, and Ollama. The example configuration uses gpt-6.1-sol and a Responses default. OpenAI through an experimental ChatGPT subscription path using Codex OAuth, with the model set to auto or an account model slug. Gemini Deep Research via the Interactions API, where the model field is an agent identifier rather than a standard model name. Tavily Research, with reasoning depth set to mini, pro, or auto. Self-hosted Allen AI DR-Tulu and Gnosis/openJiuwen DeepSearch, both of which require a running service, and for Gnosis a shared output directory. That provider list is the main tradeoff to understand. A single tool interface reduces integration work, but provider differences still surface in configuration: API style, model or agent identifiers, reasoning effort, and service endpoints are not uniform. The OpenAI backend accepts an optional reasoning_effort, for example medium, while other backends interpret the same configuration fields