AnswerHatch MCP

Put a cite-or-refuse answer box on your docs, from the chat where you already work. Ask your agent to add cited answers to a site and this server does the whole onboarding: register the domain, prove you own it, crawl the pages, and hand back a one-line embed. No dashboard, no operator.

What AnswerHatch is

AnswerHatch answers your visitors from your own published pages, links the source on every answer, and refuses when it has no source. It never invents an API parameter, an endpoint, or a wrong command.

Every "AI on your docs" tool optimises for always answering. For documentation that is the wrong default: a confidently wrong answer is not a bad chat, it is a broken integration and a support ticket. AnswerHatch is built the other way round.

• Cite or refuse. Every answer carries the page it came from, or the widget declines and gives the reason. The refusal is a gate in the request path, not a line in a prompt that can be talked around.

• No third-party model in the path. Your docs and your users' questions are processed on AnswerHatch's own machines. They do not transit OpenAI or any other model API.

• A provenance registry. Every crawled document carries its source URL, the date it was fetched, and its hash, so a reviewer can audit exactly what the widget is allowed to say.

See it refuse a made-up question, live: https://answerhatch.com

Related MCP server: mcp-rag-bridge

Install

Requires Node 20 or newer.

Claude Code, one line:

claude mcp add answerhatch -- npx -y answerhatch-mcp

Any MCP client, config form:

{
  "mcpServers": {
    "answerhatch": {
      "command": "npx",
      "args": ["-y", "answerhatch-mcp"]
    }
  }
}

Until the npm release lands, run it from source:

git clone https://github.com/answerhatch/answerhatch-mcp
cd answerhatch-mcp
npm install
claude mcp add answerhatch -- node /absolute/path/to/answerhatch-mcp/index.js

You need an AnswerHatch account. Create one at https://answerhatch.com (self-serve, a free trial), then sign in with the answerhatch_login tool below. For non-interactive use you can instead set ANSWERHATCH_TOKEN in the server's environment and skip the login step.

Use it

In your agent, just ask:

You:    Add cited answers to docs.example.com
Agent:  (answerhatch_login, then answerhatch_create_tenant)
        Registered. Publish this TXT record to prove you own the domain:
          name:  _answerhatch.docs.example.com
          type:  TXT
          value: ah-verify-...
You:    published
Agent:  (answerhatch_start_crawl, then polls answerhatch_status)
        Live. Paste this into your site:
          <script async src="https://cdn.answerhatch.com/api/widget.js"
            data-agency="..." data-key="..." data-env="prod"></script>

The tools enforce the order the API requires (register, prove ownership, then crawl) and each one tells the agent what to do next, so the agent drives the whole flow without you touching a console.

Tools

Security

The bearer token lives in memory for the session only. It is never written to disk, never logged, and it is scrubbed from tool output even if an API response tried to echo it back.

Links

• Product and the live refusal demo: https://answerhatch.com

• Pricing: https://answerhatch.com/pricing

License

MIT. See LICENSE.