Connect an agent (MCP)

Narrato runs a remote MCP server, so the agent in your editor or terminal can answer questions from your analytics: “what were my top events last week?”, “did sign-ups drop after Tuesday’s release?”. It’s available on the Pro plan.

Connect

1. In the console, switch to the team workspace you want the agent to read, then open Connect an agent in the sidebar.
2. Create a token. Copy it right away: it’s shown once.
3. Add the server to your agent.

Claude Code:

claude mcp add --transport http narrato https://api.narrato.app/mcp \
  --header "Authorization: Bearer <your-token>"

Cursor and other clients that take a JSON config:

{
  "mcpServers": {
    "narrato": {
      "type": "http",
      "url": "https://api.narrato.app/mcp",
      "headers": { "Authorization": "Bearer <your-token>" }
    }
  }
}

The console shows both with your token filled in.

What the agent can do

Everything is read-only, and limited to the workspace the token was created in.

Tool	What it does
get_event_schema	Columns, the most frequent event names, and what your data dictionary says they mean
query	A read-only SQL query over your events (FROM events, up to 100 rows)
list_dashboards, get_dashboard	Your dashboards, with each widget’s current data
list_reports, get_report	Your saved reports, with the queries behind them
get_context	Your data dictionary from the console’s Context page

The SQL dialect is the one in Querying your data.

Tokens

• A token reads one workspace: the one active when you created it. To connect another team, switch to it and create another token.
• Tokens last 90 days. You can have up to 10 at a time.
• Revoke a token from the same page. It stops working on its next call.
• If you leave a team, your tokens for it are revoked.
• Treat a token like a password. Unlike a project’s write key, it can read your data, so keep it out of source code and shared configs.

Limits

Each workspace gets 2,000 tool calls a day on Pro (reset at midnight UTC), and tool calls share the console’s per-user query rate limit of 60 a minute. A call over a limit returns an error that says when to retry.