
Observe agents on Tor MCP server
Which MCP servers are Observe agents on Tor's own, from the official registry and from asking its own domain, whether they answer, and what they offer. A program checked each of these, nobody copied them from a product page.
read 2026-10-28
- Registry nameorg.onionlens/tor-toolsverified in the MCP registry
The registry name sits under onionlens.org. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.
- Stateanswerschecked 2026-10-28
- Tools3
- Server calls itselfonionlens
- Server version0.1.3
- Protocol version2025-06-18
On 2026-10-28 the server answered initialize and tools/list without a sign-in. First answer in 200 ms.
- Whether its tools do the product's job has not been judged yet.
Once connected
- For "Validate v3 onion addresses without network requests", the likely first call is validate_onion_address.
- Parameters are untyped or undescribed in 1 of 3 tools: check arguments before calling.
Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.
- INITE MCP score81 of 100answers 100 · definitions 78 · context 100 · use 0 · connects 100
- Answered100% of days7 of 7 daily checks
- Median first answer220 ms
- Space in an agent's context≈ 171 tokensthe definitions of all its tools (3), which an agent carries on every turn
- Tool definitionsB · 78 of 100
- Typed results0 of 3 toolsdeclared outputSchema: results come typed, read once
- Parameters without a description: in 1 of 3
- Not saying whether it only reads or can destroy: 3 of 3
What an agent can do with it
- not sure · Run a sensor to record AI agent visits
- can do · Validate v3 onion addresses without network requests
- not sure · Create labelled traces of observed agent behaviour
Each of the product's use cases, put to Jev against this server's tool list: can an agent do it with these tools?
Worth a look
None of the five warning signs in these definitions.
A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →
validate_onion_addressValidate a Tor v3 address format, version and checksum without a network request.
list_resourcesList primary-source Tor integration references and controlled retrieval fixtures.
retrieval_fixtureReturn a deterministic UTF-8 retrieval fixture for transport and parsing checks.
How to connect
Add the server to your MCP client config. Claude Desktop, Cursor and most other clients read this shape:
{
"mcpServers": {
"tor-tools": {
"type": "http",
"url": "https://onionlens.org/mcp"
}
}
}Add it
Claude Code, in a terminal
claude mcp add --transport http onionlens https://onionlens.org/mcpclaude.ai, ChatGPT and other clients: add https://onionlens.org/mcp as a custom connector in the settings.
OpenAPI spec
3 paths, read 2026-10-28. Open the spec ↗
Each line is something a program found when it went looking: the registry entry, the server's own answer, a spec that parses. A missing line means we did not find it, not that it does not exist.
Before we count this visit
Analytics and ad measurement tell us which pages and which of our ads do their job: Google Analytics and Google Ads, the Meta pixel, Yandex Metrica, and one first-party id of our own. Nothing is stored until you answer, and declining changes nothing about the site.