joly0/gptr-mcp

•Updated 8 days ago

MCP server for GPT Researcher: deep web research with cited sources for AI assistants

Image

1.6K

joly0/gptr-mcp repository overview

⁠GPT Researcher MCP Server

An MCP⁠ server that gives AI assistants (Claude, Cursor, and any other MCP client) deep web research through GPT Researcher⁠. Instead of returning raw search results, it explores and validates many sources and hands back focused, cited research context or a full report.

This image is built from Joly0/gptr-mcp⁠ (branch combined), a fork of the official assafelovic/gptr-mcp⁠ that adds:

Streamable HTTP transport at /mcp, replacing the deprecated SSE transport
Smaller multi-stage image with optional extra pip packages at startup
gpt-researcher 0.16 on Python 3.12, including the new context filter⁠

⁠Quick start

docker run -d --name gptr-mcp -p 8000:8000 \
  -e OPENAI_API_KEY=your-openai-key \
  -e TAVILY_API_KEY=your-tavily-key \
  joly0/gptr-mcp:latest

Then point your MCP client at http://<host>:8000/mcp, for example in Claude Code:

claude mcp add --transport http gptr-mcp http://<host>:8000/mcp

A health check is available at http://<host>:8000/health.

⁠Tools

Tool	What it does
`deep_research`	Full web research on a query; returns the research context, sources and a `research_id`
`quick_search`	Fast web search returning results with snippets
`write_report`	Writes a report from a previous `research_id`, with an optional custom prompt
`get_research_sources`	Returns the sources of a previous research
`get_research_context`	Returns the full context of a previous research

It also exposes a research://{topic} resource and a research_query prompt.

⁠Environment variables

Variable	Required	Description
`OPENAI_API_KEY`	yes	API key for the LLM provider
`TAVILY_API_KEY`	yes (default retriever)	API key for Tavily web search
`CONTEXT_FILTER`	no	How scraped passages are selected: `auto` (default), `jev`, `keyword`, `embeddings` or `none`
`TYPESAFE_API_KEY`	no	Enables the Jev usefulness filter when `CONTEXT_FILTER=auto`; without it, local keyword ranking is used
`EXTRA_PIP_PACKAGES`	no	Comma-separated pip packages installed at container start, e.g. for retrievers with extra dependencies

All other GPT Researcher settings⁠ (STRATEGIC_LLM, SMART_LLM, RETRIEVER, MAX_ITERATIONS, ...) can be passed as environment variables as well. Inside Docker the server always uses the streamable HTTP transport on port 8000.

⁠Tags

latest: newest build
gptr-<version>: pinned to the bundled gpt-researcher version, e.g. gptr-0.16.1

⁠Building it yourself

git clone -b combined https://github.com/Joly0/gptr-mcp.git
cd gptr-mcp
docker build -t gptr-mcp .

Build arguments: BUILD_FROM_SOURCE=true installs gpt-researcher from GitHub instead of PyPI, with GPT_RESEARCHER_OWNER (default assafelovic) and GPT_RESEARCHER_REF (default main) choosing the repository and branch.

⁠License

MIT, see the repository⁠.

Tag summary

Recent tags

Content type

Image

Digest

sha256:8236aa477…

Size

634.8 MB

Last updated

8 days ago

docker pull joly0/gptr-mcp