GPT Researcher
by assafelovic
The GPT Researcher MCP server runs deep, multi-source web research and returns a validated, cited report instead of a single raw search result.
Clone & configure
git clone https://github.com/assafelovic/gptr-mcp.git && pip install -r requirements.txt
About
gptr-mcp wraps the open-source GPT Researcher engine in an MCP server so an agent can trigger real research instead of a one-shot web search. Its deep_research tool crawls and cross-checks dozens of sources before it settles on what's reliable, while quick_search trades that depth for speed when a fast snippet-level answer is enough. write_report then compiles whatever was found into a structured report, and get_research_sources / get_research_context let the agent inspect the evidence trail behind any conclusion rather than taking it on faith.
It ships as a plain STDIO server for local clients like Claude Desktop, or as an SSE/HTTP service behind the included Docker Compose setup for team or web deployments. Either way it needs its own OpenAI and Tavily API keys configured separately from whatever model is driving the conversation, since the research itself runs as an independent multi-step process.
Key features
- deep_research explores and validates multiple sources before drawing conclusions, instead of returning raw hits
- quick_search offers a faster, lower-depth alternative when speed matters more than exhaustive coverage
- write_report compiles findings into a structured, readable report on demand
- get_research_sources and get_research_context expose the underlying evidence behind any finding
- Runs as STDIO, SSE, or a Dockerized HTTP service depending on how it's deployed
- Built directly on the open-source GPT Researcher engine rather than a thin search wrapper
Use cases
- Compiling a cited background report on a company, market, or technology before a meeting
- Running competitive or literature research that needs multiple corroborating sources, not one link
- Feeding a downstream writing task with a research_resource pull instead of manual browsing
- Standing up a shared research endpoint for a team via the Docker/SSE deployment
Available tools
deep_research
Conducts thorough web research on a topic, cross-validating multiple sources before settling on what's relevant and reliable.
quick_search
Runs a fast web search that prioritizes speed over exhaustive validation, returning results with snippets.
write_report
Compiles prior research findings into a structured, written report.
get_research_sources
Returns the list of sources that were used to produce a given research result.
get_research_context
Retrieves the full research context and intermediate findings behind a task.
Frequently asked questions
Does it need its own API keys?
Yes. You configure an OpenAI key and a Tavily key (for search) in claude_desktop_config.json or the server's environment, separate from whatever model is running the conversation.
How is this different from a plain web search MCP tool?
deep_research cross-checks findings across many sources before it reports anything, rather than handing back raw, unverified search hits.