Your agent already knows your Git. Let it know your head.
Everything you know — in one place, in any form. Your agent writes it into one map from its own session, so you don't have to. You see the map as a canvas of your notes, joined by connections.
Works with Claude Code, Cursor, Claude Desktop, claude.ai and any MCP client.
while you work
Monday, 10:00. Where did I stop on each project?
Open the map: every project is on it, in blocks, each a title and its text. Your agent wrote them between commits, whenever you asked it to.
Your agent, over MCP
Claude Code or Cursor reads the map and writes to it without leaving its session.
In claude.ai or Claude Desktop, connect in one click. For Claude Code, paste an MCP key from Settings → MCP Server into this command:
claude mcp add --transport http corlecti-map https://map.corlecti.com/api/mcp \
--header "Authorization: Bearer mmap_live_YOUR_KEY"Telegram
Send the bot a line or a voice note from the train. It becomes a block on the map you name.
The browser
Dictate to the agent built into each map, in the map's chat. From the dashboard, search every map for a word and see its line.
A map can also start from a PDF, a web page or the agent interviewing you. How the agent builds a map
The agent lays a map out by your rules.
The agent sends the blocks. Your settings decide where they sit and what colour they are.
while you're away
In the morning the map has what it didn't have last night.
Give a map a schedule and an instruction, and the map's chat works on it while you sleep. And the bot writes first: one question about two blocks that look related, and your answer becomes the connection. Its questions use no model, so they spend nothing, and /reminders off stops them.
You wrote about this in March.
Open a block and unfold Nearby, its list of what is close in meaning, on this map and your others. Two blocks can be close with no word in common. Press Link beside a block from this map to connect the two.
The graph proposes a missing connection as a dotted line. You press Link to make it. What the map notices
control
Your agent, your rules.
An MCP key narrowed to one map or folder
The agent that holds it sees nothing else.
A snapshot before every agent edit
Your agent over MCP and the map's chat get one. Roll back from History, and the rollback itself can be undone.
Your model, and how long it thinks
Set both for the map's chat and each Telegram bot; your agent over MCP keeps its own. On your own provider key from Anthropic, OpenAI, Gemini or DeepSeek, your provider bills that work.
One read of the map tells your agent where each project stopped. Rescanning every repository costs far more.
I run eleven projects through one map.
- Pending: home.control.launch
- Roadmap
- Pending: home.control.principles
Run a Telegram bot under your own name, and let the people you choose write to it. What writes to your map
views
One map. Seven ways to hand it over.
Canvas
Your own layout, and maps inside maps.
Graph
The same map as a network of connections.
Document
Markdown, with numbered sections and cross-references.
PDF
Fonts embedded, ready to print or email.
Slides
A PowerPoint deck, one slide per block.
Link
Read-only, no account needed, gone when you delete it.
.mmap
A plain JSON file, for backups and your own scripts.
Take it all away in one .mmap file — blocks, connections, layout. On the free plan too.
A month later, a block is still where you left it. Every view in detail
Start free.
- The free plan has the map's chat, MCP, the bot and 10 maps.
- Its AI allowance is how much AI work you can ask for this week: about 1230 blocks the map's chat writes. Your own provider key leaves it untouched.
Your first sign-in adds starter maps for your line of work.
How is this different from Obsidian, Notion, or an AI mind-map generator?
What if I just ask Claude?
Do my notes go into the model?
From here — decide for yourself.