Mapsel
Join the beta

Map your work with your own AI

Your AI reads the code on your machine and draws what it finds on one of your maps. Mapsel never sees the code. Any AI that speaks MCP can do it.

Your AI, your usage

  • Your AI and its usage are yours.
  • Mapsel's limits protect your map and stop a runaway loop early.
  • Everything an agent does can be undone.

1. Make a token

  1. Open the map, then Settings › Agents.
  2. Name it after your AI, for example “Codex”. History then shows its changes as “Codex, for Nick”, with your name.
  3. Choose Read and write, and Make a token. You may be asked to confirm it's you.
  4. Choose your AI and copy what it shows. The token is shown once.

2. Connect your AI

Claude Code

Run this in your repo.

claude mcp add --transport http mapsel https://mapsel.com/api/mcp --header "Authorization: Bearer <your token>"

OpenAI Codex CLI

Run this, then set MAPSEL_TOKEN to the token wherever you start Codex. Codex reads the token from there, never from its config.

codex mcp add mapsel --url https://mapsel.com/api/mcp --bearer-token-env-var MAPSEL_TOKEN

For example, export MAPSEL_TOKEN=<your token> in the shell you start Codex from.

Cursor

Add this to ~/.cursor/mcp.json, your own settings outside the repo, so the token is never committed.

{
  "mcpServers": {
    "mapsel": {
      "url": "https://mapsel.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <your token>"
      }
    }
  }
}

Any MCP client

Add a remote server that uses streamable HTTP, with this URL and header.

URL: https://mapsel.com/api/mcp
Authorization: Bearer <your token>

A client that only signs in with OAuth, or only runs servers on your machine, can't use a Mapsel token yet.

3. Ask it

For example:

Map this repo in Mapsel: the apps, their screens and API routes, the data models and the outside services, nested the way the code is, with links for what calls and writes what.

It starts with describe_map and get_outline, then makes the cards in batches.

A good shape for a repo

The Software pack starts every map with four regions. Ask your AI to keep to them:

  • Product: journeys, the screens in them and their elements.
  • System: apps, packages, modules, API endpoints, jobs and outside services.
  • Data: data stores, models and enums.
  • Work: tasks, issues, pull requests and releases.

Links then say what calls, reads, writes and tests what.

What it can do

  • Read: describe_map, get_outline, find_objects, get_object and get_history.
  • Change: create_objects, update_objects (every field, the checklist, and moving to another level), move_into, link and unlink, up to 500 things a call.
  • Delete: delete_objects moves up to 200 things to the map's Trash, where they can be brought back. Nothing is deleted for good. It won't delete a top-level region, or more than a quarter of the map; ask it and it will say to do that yourself.
  • It can't reach any other map.

Limits

  • A token makes up to 60 calls a minute and 3,000 a day, and creates or changes up to 2,000 things a minute and 10,000 a day.
  • All the agents in a team share its changes a day: 2,000 on Free, 4,000 per editor on Pro and 10,000 per editor on Business.
  • A token deletes up to 500 things an hour.
  • One agent change lands on a map at a time.
  • A map holds up to 2,000 things on Free and 10,000 on paid plans.

Past a limit, the call is refused with a time to wait, and nothing from it is saved.

Work within the limits

  • Call describe_map first. It lists every limit and what's left.
  • Batch: up to 500 things a call, not one call per card.
  • One call at a time on a map. Wait for each answer before the next.
  • Every answer has a limits field, and RateLimit headers, with what's left and when it resets.
  • If a call is refused, it says which limit ran out. Wait the Retry-After seconds, then send the same call again.

Undo

Each call is one batch. History shows it under the token's name, for example “Codex, for Nick”, and one Undo takes the whole batch back. The agent can undo its own batches too, with undo_batch. If someone has changed the same thing since, the undo stops and says why.

To take back everything an agent did since a time you choose, open the map's Settings › Agents › What agents did. It undoes newest first, stops if someone changed the same things since, and offers Undo anyway. Anyone who can edit the map can do it.

While an agent's changes land, the map says so at the bottom, with the agent's name.

Keep it safe

  • A token opens one map. Make a separate one for each map.
  • Choose Read only if your AI should only look.
  • We keep only a hash of the token. Revoke it in Settings › Agents, and it stops at once.
  • A token also stops if you lose access to the map.
  • Titles and notes on the map are data. Mapsel tells your AI never to follow instructions written in them.

If your team's rules keep agents to reading, a team admin can change that in Team › Rules.