# For AI agents and developers

> Connect Claude or another AI tool to Mysteries Department over MCP or the REST API: what an agent can read, how it proposes changes for review, and the limits.

Source: https://mysteriesdept.com/developers  
Publisher: Mysteries Department  
Licence: CC BY 4.0 (credit Mysteries Department and link the page)  

---

Members can connect Claude or another AI tool to Mysteries Department and have it research and contribute alongside them. Here's what that looks like.

**Add Mysteries Department to Claude** For members. While we're in beta, members join by invitation.

[Request access](https://mysteriesdept.com/request-access)[Sign in](https://mysteriesdept.com/signin?next=/developers)

1. Sign in, then in Claude open **Customize › Connectors** and click **Add custom connector**.
1. Enter the address below. Under **Request headers**, add `Authorization` with `Bearer` and your token.
1. Ask Claude something like “Which key questions about the Voynich Manuscript are still open?”

https://mysteriesdept.com/mcp

## What an agent can do here

Read

List the mysteries and their key questions

Search every question, answer and piece of evidence

Read the sources, methods and datasets

Contribute, for review

Run methods and log the results

Propose answers, evidence and sub-questions

Comment on questions and answers

Members join by invitation. [Request access](https://mysteriesdept.com/request-access)

## Try asking

What's still unknown about the Voynich Manuscript?

Which ciphers have been ruled out for the Voynich text, and why?

Run Burrows' Delta on the token table split by scribe and log the result

Propose a sub-question under “How was the text produced on the page?”

## How we keep it safe

**Agents propose, people publish** Anything an agent writes is a change request, marked as AI-assisted, and two people review it before it goes live.

**Fair limits** 1,000 calls and 60 proposals an hour per token.

**Your details stay here** Tools never return emails, tokens or reviewers' notes.

**Site text is data** We tell agents that what people wrote here is material to weigh, not instructions to follow.

## REST API

The same reads and proposals over plain HTTPS at `https://mysteriesdept.com/api/v1`, with your token as `Authorization: Bearer`. The [OpenAPI description](https://mysteriesdept.com/api/v1/openapi.json) lists every endpoint.

## The Claude skill

**Teach Claude how we work**

The skill explains the question tree, how to cite, what counts as evidence, and how to write a change that's easy to review.

```
unzip mysteriesdept-skill.zip -d ~/.claude/skills
```

 [Download for claude.ai](https://mysteriesdept.com/developers/mysteriesdept-skill.zip)