Your brain is not locked inside Nisa. Any agent that speaks MCP — Claude Code, Claude Desktop, Cursor, Langdock, your own script — can search it, read a subject, and write facts back.
What you need
-
A personal access token. Settings → Developer → Personal access tokens → create one. The value is shown once, so copy it before you close the dialog.
-
The server address, which is the same for everyone:
https://api.dramalabs.co/mcp-tools/serve
The token is how the server knows who you are. It acts as you: it reads what you may read, writes as you, and never sees a page your account is not allowed to see.
Treat the token like your own password
A token is an account key, not a brain key. It opens the brain, and it also opens the other things your account can reach: your conversations with Nisa, the names of your saved secrets, and creating and running Skills. It cannot reach another company's data, and it cannot do the things reserved for an owner unless you are one.
So: give a token to a tool you run, never to a person. One token per tool, named for the tool, so you can revoke one without breaking the others. If you want somebody else's agent to work with your brain, give them their own account and let them make their own token, rather than sending them yours.
Add it to your agent
Claude Code, one command:
claude mcp add --transport http dramalabs https://api.dramalabs.co/mcp-tools/serve --header "Authorization: Bearer <your token>"
Claude Desktop and Cursor, in the MCP section of the settings file:
{
"mcpServers": {
"dramalabs": {
"url": "https://api.dramalabs.co/mcp-tools/serve",
"headers": { "Authorization": "Bearer <your token>" }
}
}
}
Langdock: add an MCP integration, use the address above, and send the token as an
Authorization: Bearer header.
Settings → Developer → Connect an agent has the same snippets with a copy button.
What the agent can do
| Ask for | It does |
|---|---|
| Something about the company | Searches the brain the way Nisa does, and gets the same ranked results with dates and sources |
| A person, client or project | Opens that subject's page: what is compiled, the facts, the goals, the open questions |
| A list of facts | Returns the live ones, filtered by kind or by a word |
| To record something | Writes one keyed fact through the same checks Nisa's writes go through |
| To remove something | Retracts a fact, which leaves a dated note rather than a hole |
| A picture of the brain | Counts, what changed last night, and what wants a look |
What it cannot do
It cannot see past your own access. If a page is restricted to a group you are not in, it is absent from every answer — not refused, absent.
It cannot pretend to be someone else. Every fact it writes records which agent wrote it, so you can tell later what came from Nisa, from a Skill, or from your own client.
A guest account cannot write. Neither can a workspace token. Writing needs a personal token belonging to a member.
If something goes wrong
The agent lists no tools. The token is wrong, expired, or was revoked. Create a new one.
A search returns nothing you expect. The brain may genuinely not hold it, or the pages are restricted to a group you are not in. Open the brain in the dashboard and search the same words there — what you see there is exactly what the agent sees.
A write is refused. The message says why. The usual reasons are a value that looks like a credential, which never goes in the brain, and a fact the workspace's refusal list excludes.
Losing a token
Revoke it in Settings → Developer. Access stops at once. Nothing it wrote is removed, because that is in your brain where you can read and change it yourself.