
Robin Dark Web OSINT MCP Server: Setup for Claude Code and Codex
Lets Claude Code and Codex search and read dark web sites over Tor, and save investigation reports.
Robin is a dark web OSINT app that also runs as an MCP server in Docker. Your assistant refines a query, searches sixteen onion search engines over Tor, filters the results and reads pages, then writes a report. Tor runs inside the image, and the README says your assistant's own model does the reasoning.
The README says it is for lawful investigations and that accessing some dark web content may be illegal where you live. Page text comes back as untrusted data, so do not let the assistant act on instructions found there.
Good for
- Security researchers and analysts who run OSINT investigations on dark web sources
- Threat intelligence teams checking for leaked data or mentions of their brand
- Developers who want a dark web search tool inside Claude Code or Codex
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install Robin: AI-Powered Dark Web OSINT MCP
Claude Code
- 1
Copy this prompt.
Add the Robin: AI-Powered Dark Web OSINT MCP server to Claude Code. Run this command: claude mcp add robin -- docker run -i --rm -v robin-investigations:/app/investigations apurvsg/robin mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Docker.
Read-only mode: There is no read-only mode. The only thing it writes is saved reports in the investigations volume.
Install Docker and pull the image first (docker pull apurvsg/robin:latest) so the first start is not also a download. Tor is bundled in the image. The config mounts a named volume so saved reports persist and show up in the Robin web UI. A Tor investigation takes minutes, so the README suggests raising the tool timeout (about 600 seconds) in hosts such as Codex that stop long calls early. No API key is needed in this mode, because Robin uses your assistant's own model when the client supports it; a model key in .env is only for Robin's own web UI.
From the project README. Source.
Tools
Runs a whole dark web investigation in one call
Refines your question, searches, filters, reads pages and writes a summary, then saves it. The source calls it a convenience that uses your assistant's model if it supports sampling.
Try: “Investigate leaked credentials for example.com on the dark web.”
Write ·
robin_investigateTurns your question into a short dark web search
Rewrites a question as a search query of five words or fewer. It uses your model when the client offers it.
Try: “Refine this question into a dark web search query.”
Read ·
robin_refineSearches sixteen onion search engines over Tor
Searches onion search engines for one refined query and returns the results with short IDs. The titles come from the sites found, so treat them as untrusted.
Try: “Search the dark web for "example.com credentials dump".”
Network ·
robin_searchKeeps only the on-topic results
Judges every result of your latest search for relevance and returns the ones that fit, most relevant first.
Try: “Filter those results down to what is on topic.”
Read ·
robin_filterReads pages over Tor
Reads the pages of result IDs from the filter step, or onion addresses you give, over Tor and returns the text as untrusted data. Non-onion hosts are refused unless you allow clear web.
Try: “Read the pages for the filtered results.”
Network ·
robin_scrapeSaves a report you wrote
Saves the summary you wrote from search and scrape results into the investigations folder. Robin picks the file name.
Try: “Save this summary as an investigation report.”
Write ·
robin_save_investigationLists saved investigations
Lists saved reports on this machine, newest first.
Try: “Show my saved investigations.”
Files ·
robin_list_investigationsLists models Robin can run on its own
Lists models available to Robin from configured providers. An empty list is normal, since your assistant's model can be used instead.
Try: “Which models can Robin use?”
Read ·
robin_list_modelsChecks that Tor is up
Reports whether Tor is running and can also probe a model or the search engines when you ask.
Try: “Is Robin ready? Check Tor.”
Read ·
robin_health
Read from the source code
Permissions
medium risk · 4 things to watch
Medium risk
The assistant reads untrusted dark web pages and saves reports to a Docker volume. Treat what it reads as data, check the law where you live, and keep other tools away from this session.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
Dark web pages are untrusted and may try to steer the assistant
Robin fences scraped text and strips hidden characters. Still, do not give the same session tools that can send data or run code. github.com
Accessing some content may be illegal
The README says legality depends on your jurisdiction and puts responsibility on you. Check local law before you use it. github.com
Reports are saved to disk
Saved reports go into the investigations volume. The README says the agent never picks the file path. github.com
Page reads go through Tor
The README says searches and page scrapes go through Tor, while model calls go directly. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode is documented. The only write is saving reports into the investigations volume. |
|---|---|
| Talks to | Tor network (search engines and onion sites). Your model provider, if you configure Robin's own model |
| Sign-in and keys | No account. A model API key goes in a .env file only for Robin's own web UI or its own model. |
Questions
Is Robin an official server?
No. It is a community project by the Robin author that ships its own MCP mode. It is not connected to any dark web service.
Do I need Docker?
Yes for the documented setup. The config runs the apurvsg/robin image, which includes Tor, so you do not install Tor yourself.
Do I need an API key?
Not for MCP mode. The README says Robin uses your assistant's own model. A model key in .env is only for the web UI or Robin's own model.
Does it work in ChatGPT chat?
No. The README says ordinary ChatGPT chat cannot run it, because chat uses connectors on OpenAI's servers. It works in Codex and the ChatGPT desktop app's Codex host.
Why does the first call time out?
A Tor investigation takes minutes, which can pass a client's default tool timeout. The README suggests raising it to about 600 seconds and pulling the image first.
Is it legal to use?
That depends on where you live. The README says accessing some dark web content may be illegal and the author takes no responsibility for misuse.
Where are my reports saved?
In a Docker volume called robin-investigations. The Robin web UI shows the same reports if it mounts that volume.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
apurvsg/robin- Registry
- Docker
- Transport
- stdio
- Version
- 3.1.0
- License
- MIT
- Last commit
- Oct 3, 2026
- Open issues
- 1
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- Robin README · github.com
- Robin MCP server source · github.com