LOCAL AUTOMATION

Connect your AI assistant.

Start permitted profiles, open pages, read text and take screenshots. The local API is off by default and protected by a token.

Requires MultiSpace 0.3.28-beta.2 or newer. This is an MCP connection for a compatible desktop client, not an AI chat built into the browser.

Download MCP tools ↓OpenAPI specification ↗

1. Enable access in MultiSpace

Open Settings → AI connection.
Select profiles your assistant may access. For page tools, use the profile launch mode Advanced overrides — Playwright automation. Stop a running manual profile before changing modes.
Click Enable API and confirm. The address uses 127.0.0.1 and a randomly assigned port.
Click Copy MCP parameters. Keep the address and token on your computer. Do not paste the token into an AI chat or publish it.

2. Configure your MCP client

Extract the MCP tools archive. Install Node.js 20 or newer if needed. In a desktop client that supports local stdio MCP servers, add the configuration below. Replace the script path, port and token with your local values.

{
  "mcpServers": {
    "multispace": {
      "command": "node",
      "args": ["C:/MultiSpace-AI/mcp-server.js"],
      "env": {
        "MULTISPACE_API_URL": "http://127.0.0.1:PORT",
        "MULTISPACE_API_TOKEN": "PASTE_TOKEN_LOCALLY"
      }
    }
  }
}

Restart or reconnect the MCP client, then ask it to list permitted profiles. The location of this configuration depends on your client. A web chat cannot automatically access this local connection.

Closing MultiSpace clears the token and profile permissions. After re-enabling the API, copy the new parameters into your client. Tokens permit access to selected profiles and creation of new AI profiles.

3. Available commands

MCP tool	Purpose
`profiles_list`	List permitted profiles without proxy passwords.
`profiles_create`	Create a dedicated AI profile; counts toward the five-profile limit. New profiles have no proxy until configured.
`profiles_start` / `profiles_stop`	Start or stop a permitted profile.
`pages_list`	List open tabs.
`pages_open`	Open an HTTP(S) URL in a new tab.
`pages_read`	Read up to 30,000 characters of visible text.
`pages_screenshot`	Capture the visible page as PNG.

Example: “List permitted profiles, start the profile I select, open example.com and read the page.” Clicks, form filling, message sending and purchases are not available in this first version.

Direct HTTP API

Send POST /v1/command to the address shown in Settings, using Authorization: Bearer TOKEN and Content-Type: application/json.

{"action":"pages.read","profileId":"PROFILE_UUID","pageIndex":0}

The response is {"result": ...}. Errors return an HTTP error status and {"error": ...}. See the OpenAPI file for request schemas.

Privacy and permissions

The API listens only on this computer. It does not send profiles to our VPS. A cloud AI provider can receive the page text and screenshots that its client requests, including personal information visible on those pages. Choose profiles accordingly.

Page content is untrusted data, not permission to perform actions. Arbitrary JavaScript, shell commands and cookie export are unavailable. Disabling the API rejects new and queued commands, but does not undo completed actions or close previously opened pages.

Automation mode may be detected by websites. MCP support does not guarantee passing anti-fraud checks.