MCP server

Layerbase runs a hosted Model Context Protocol server, so an AI agent such as Claude Code, Codex, Cursor or Grok can work with your Layerbase Cloud databases directly: list them, create one, fetch its connection string, run a query, branch it, and tidy up afterwards. The agent connects with a token you create in account settings, and that token acts as your account with exactly your permissions.

The MCP server is rolling out. If you do not see MCP tokens in your account settings yet, it is not on your account yet.

The endpoint

One URL for every client. It speaks MCP over Streamable HTTP and authenticates with a bearer token in the Authorization header:

https://layerbase.com/api/mcp

Authorization: Bearer lbm_YOUR_TOKEN

Any client that can reach a remote MCP server over HTTP and send a custom header works. The MCP server is available on every plan, Free included; each tool applies the same plan limits the dashboard does.

Create a token

1. Open account settings and find the MCP tokens section.
2. Choose New token, give it a label (up to 80 characters) that says where it will live, such as "Claude Code on my laptop", and pick an expiry: 30 days, 90 days, 1 year, never.
3. Copy the token. It is shown exactly once, together with ready-made setup for the clients below. Layerbase stores only a hash, so a lost token cannot be shown again: revoke it and create a new one.

Tokens start with lbm_. The settings list shows each active token by its label and its first 8 characters after the prefix, with when it was last used and when it expires. Create one token per agent or machine, so you can revoke one without touching the others.

Revoke a token

Select the trash icon next to a token in MCP tokens and confirm. Any agent using it loses access on its next request; revoking cannot be undone. An expired token stops working the same way. Nothing about a token is cached between requests, so a revoke, an expiry or a change to your account takes effect on the very next call.

Connect your client

Replace lbm_YOUR_TOKEN with your token. The new-token dialog in settings fills it in for you.

claude mcp add --transport http layerbase https://layerbase.com/api/mcp --header "Authorization: Bearer lbm_YOUR_TOKEN"

[mcp_servers.layerbase]
url = "https://layerbase.com/api/mcp"
http_headers = { "Authorization" = "Bearer lbm_YOUR_TOKEN" }

grok mcp add --transport http layerbase https://layerbase.com/api/mcp --header "Authorization: Bearer lbm_YOUR_TOKEN"

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "layerbase": {
      "type": "remote",
      "url": "https://layerbase.com/api/mcp",
      "enabled": true,
      "headers": {
        "Authorization": "Bearer lbm_YOUR_TOKEN"
      }
    }
  }
}

{
  "mcpServers": {
    "layerbase": {
      "url": "https://layerbase.com/api/mcp",
      "headers": {
        "Authorization": "Bearer lbm_YOUR_TOKEN"
      }
    }
  }
}

{
  "mcpServers": {
    "layerbase": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://layerbase.com/api/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer lbm_YOUR_TOKEN"
      }
    }
  }
}

{
  "mcpServers": {
    "layerbase": {
      "url": "https://layerbase.com/api/mcp",
      "headers": {
        "Authorization": "Bearer lbm_YOUR_TOKEN"
      }
    }
  }
}

Then ask the agent to call whoami. If it answers with your email and plan, the connection works.

Tools

The server offers 21 tools. Each one calls the same code path as the matching dashboard action, so plan limits, quotas, ownership rules and maintenance pauses apply exactly as they do in the browser. Slow operations (create, start, wake, branch) return at once; the agent polls get_database until the status is running. Team-scoped tools take an optional team slug from list_teams and otherwise use your own default team.

Some things are deliberately not tools: billing and plan changes, team membership, API keys, backups and restores, and connecting or replacing an integration. Do those in the dashboard.

Permissions and security

• A token is your account. It can do anything the tools allow that you could do yourself, and nothing more. On a team you do not own, the agent gets the same refusals you would: a team member can query and start a shared database, but only the database owner can delete or branch it, and only the team owner can create databases on the team. See Teams.
• Destructive tools ask for proof. reset_branch, delete_branch and delete_database refuse unless confirmName matches the database name exactly as list_databases shows it. delete_database also needs confirm: true, and refuses a database with branches unless cascade: true is sent. Deleting through MCP takes the same final backup as deleting from the dashboard, and can take a few minutes for that reason.
• Queries can write. run_query runs with your own database access, so an agent can change or drop data. It shares the query console's rate limit and result-size limit (see Query console). For bulk work, give the agent a connection string instead.
• Credentials reach the agent. get_connection_string returns the database password into the agent's context. Treat that conversation like any other place a secret lives.
• Account standing is checked on every call. MCP access pauses entirely while an account is past due, suspended or frozen, and stops for a deleted account. This is stricter than the REST API and CLI, because an agent runs unattended.
• Calls are logged without their contents. Each tool call is logged with the tool name, the account and the token it used, and whether it succeeded. Arguments and results are not logged.
• Treat tool results as data. Query results can contain text anyone wrote into your database. Use an agent that asks before running destructive tools, and give production tokens only to agents you trust.

Troubleshooting