Atlas

Connect an AI assistant (MCP)

Atlas publishes your governed Knowledge Base as a read-only MCP server, so an outside AI assistant — Claude Desktop, Claude Code, or any MCP-capable client — can search and read your definitions of record, notes, and reference docs while it works. It never writes, and it only ever sees what you can.

The connection URL

Point your MCP client at this address — the same one for everybody. You pick your organization while you connect:

https://atlas.rehearsal.datacov.es/mcp

New to this? See the step-by-step setup for your assistant

In most clients you add this as a "remote" or "HTTP" MCP server. The client discovers how to sign in from the URL itself — you don't configure anything else by hand.

Two ways to authenticate

The outlet accepts a bearer token that grants read-only access to your knowledge. You can get one either way — the connection behaves identically.

Sign in (OAuth)

Recommended

Best when your client supports MCP sign-in (Claude Desktop does). Add the URL and the client walks you to an Atlas approval screen — no token to copy. Each connected tool is listed separately and you can revoke any one of them at any time.

Paste a token (PAT)

Static token

For clients that just want a bearer token. Create a personal access token in Settings, copy the atlas_pat_… value once, and paste it into your client. It carries read-only knowledge access and you can revoke it whenever you like.

Quick setup

Most clients add a remote MCP server in one step. Two common ones are below — the URL is already yours; swap in a token from above only if your client doesn't do the sign-in flow.

Claude Code

Add it as a remote (HTTP) server, then run /mcp to sign in — or pass a token directly.

Sign in with OAuth 
claude mcp add --transport http atlas https://atlas.rehearsal.datacov.es/mcp
…or connect with a personal access token 
claude mcp add --transport http atlas https://atlas.rehearsal.datacov.es/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

GitHub Copilot (VS Code)

Add the server to .vscode/mcp.json (or run "MCP: Add Server" from the Command Palette). VS Code handles the sign-in.

.vscode/mcp.json 
{
  "servers": {
    "atlas": {
      "type": "http",
      "url": "https://atlas.rehearsal.datacov.es/mcp"
    }
  }
}

Make it your default source of truth

Atlas already tells the assistant, when it connects, to treat your Knowledge Base as the source of truth. The connection makes that knowledge available — it can't force the assistant to consult it. To make it a habit on every question, add this rule to your client's instructions:

Assistant rule 
Atlas is our organization's governed source of truth for data
definitions, metrics, and reference docs. Before answering any data or
analytics question, search Atlas (search_knowledge / get_page) and ground
your answer in what you find. Prefer Atlas definitions over assumptions,
and say so when Atlas has no answer.
  • Claude Code — add it to your project's CLAUDE.md.
  • GitHub Copilot — add it to .github/copilot-instructions.md.
  • Claude Desktop and other clients — paste it into the project's custom instructions.

What the assistant can do

The outlet exposes a small set of read-only tools. The assistant calls them itself, as needed, to find and read the right knowledge before it answers.

  • list_workspaces List the workspaces you can reach.
  • list_areas List the areas inside a workspace, to narrow a search.
  • get_page_tree Show a workspace's knowledge base as an indented map of its pages.
  • search_knowledge Semantic search across governed definitions, reference docs, and notes.
  • get_page Fetch the full body of a single page by id or slug.
  • get_history Show a page's version history, with a diff of the two most recent versions.
  • fetch ChatGPT deep-research compatibility: read one page by the id a search returned.

How the assistant uses them

Given a question, the model chains these read-only calls to ground its answer in your agreed knowledge — a typical retrieval looks like this.

  1. 1

    Pick a workspace with list_workspaces, and optionally narrow to an area with list_areas.

  2. 2

    Run search_knowledge with the question to find the most relevant governed pages.

  3. 3

    Open the best matches in full with get_page.

  4. 4

    Follow the trail — list_related, list_links, and get_history — to gather supporting context and see how a definition evolved.

  5. 5

    Answer, grounded in your organization's agreed knowledge, pointing back at the pages it used.

What it can — and can't — see

  • Read-only. It can search and read governed Knowledge pages — never write, delete, run a query, or reach your warehouse data.
  • Scoped to what you can reach. It sees exactly the pages you can; a page you cannot read simply looks not-found, with no hint that it exists.
  • No secrets. Passwords, keys, and connection configuration never leave Atlas.
  • Your organization only. A token is bound to your workspace's address, so it can never reach another organization's knowledge.