Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@gptr-mcp run deep research on the impact of AI on healthcare"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Vendored, patched gptr-mcp — the Model Context Protocol server for GPT Researcher — as deployed on LinuxServer2 for Open WebUI at ronzz.org.
This is a backup of the patched source, not a GitHub fork. Upstream is
tracked manually; the diff against upstream master is listed below.
Upstream docs are preserved at docs/upstream-README.md.
The official gptresearcher/gpt-researcher image is amd64-only, and the
full GPT Researcher app has no hosted service. LinuxServer2 is aarch64, so
the MCP server is built from source (python:3.11-slim, arch-agnostic) with
the fixes below.
Related MCP server: Google Search MCP Server
File | Change | Reason |
| Docker auto-detection no longer overrides an explicit | Upstream forces SSE inside Docker; Open WebUI supports Streamable HTTP only |
| Startup no longer hard-exits without | We use the native |
| Installs runtime system libs ( | Upstream installs none — WeasyPrint / |
| Adds | gpt-researcher's metadata does not declare the retriever backends |
| New — excludes | Prevent secrets / local state baking into the image |
Source: /opt/gptr-mcp/ · container gptr-mcp (gptr-mcp:v1), 127.0.0.1:8000,
mem_limit 2g, joins the external network openwebui_default
Open WebUI connection: id gptr → http://gptr-mcp:8000/mcp (Streamable HTTP, auth none)
Full runbook (gated): https://wikibase.ronzz.org/wiki/RonzzIT:Runbook/OpenWebUI
Registration script: deploy/register-gptrmcp.sh
cd /opt/gptr-mcp
docker compose up -d # recreate after any .env change
docker compose build && docker compose up -d # after editing source/requirements.txtCopy .env.example to .env (mode 0600) and fill the keys.
CONTEXT_FILTER=keyword selects BM25 and needs no embedding provider; a
non-empty OPENAI_API_KEY is still required because GPT Researcher builds the
embedding client eagerly at startup (it is never called in keyword mode).
deep_research, quick_search, write_report, get_research_sources,
get_research_context.
MIT — see LICENSE. Upstream © Assaf Elovic and contributors.
This server cannot be deployed
Agent-driven search: build, import, tune, search, and score result quality — all over MCP.
Make your knowledge agent-ready. One MCP endpoint, 5 connectors, 3 search modes.
Live AI-native web search with citations. One tool for every MCP client. Flat per-request pricing.
Enrich, search, assess, and manage threat intelligence through 80+ typed MCP tools.
