AI access (MCP)

Every tenant exposes its telemetry over an MCP server, so your own AI tooling — Claude Code, Cursor, VS Code, Codex CLI, Gemini CLI, JetBrains — can query metrics, logs, traces, and your demand set directly. Access is authenticated with a personal access token you create in the LeanSignal app. Unlike LeanBuddy, MCP access is not metered by credits: your client brings its own model, so the model spend happens — and is capped — in your own tooling.

The endpoint

The MCP server is plain Streamable HTTP with bearer authentication, at a per-tenant host:

https://<tenant>-mcp.<region>.leansignal.io/mcp
Authorization: Bearer <token>

The exact endpoint for your tenant is shown on the Preferences page, under Access tokens, with a copy button.

Create a token

1. In the LeanSignal app, go to Preferences → Access tokens. The same tokens work for the CLI and for MCP.
2. Give the token a name that identifies where it will live — e.g. work laptop — and click Create.
3. Copy the token immediately — it is shown only once. What is stored and listed afterwards is only its prefix.
4. Pick your client under Add it to your client and copy the ready-made configuration snippet — it already contains the endpoint and your new token.

Tokens expire after 90 days. The token list shows each token’s prefix, scopes, creation date, last use, and expiry; expired tokens stay listed until you remove them.

Scopes

Tokens are read-only by default: they can query stored and agent-local telemetry, demands, dashboards, and alert rules — but change nothing.

Editors and admins can opt a token into write access at creation time:

Scope	What it allows
(default)	Read-only: query metrics, logs, traces, and the demand set.
Allow writes	The AI client can create and update demands, dashboards, alerts, rules, and channels — acting as you.
Also allow deletes	Deletes too. Deleting a demand object purges the stored telemetry only it covered, so this is gated behind its own switch.

Grant write access only to tokens used by tooling you supervise — everything a write-scoped token does is done as your user. What write access unlocks — and the guardrails around it — has its own page: Agentic operations.

Connect a client

Every MCP client that speaks Streamable HTTP needs the same two values: the endpoint URL and the Authorization: Bearer <token> header. The Preferences page generates these snippets with your values filled in; the shapes are:

Claude Code — run once in a terminal (per project, or add -s user for everywhere):

claude mcp add --transport http leansignal \
  https://<tenant>-mcp.<region>.leansignal.io/mcp \
  --header "Authorization: Bearer <token>"

Cursor — ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "leansignal": {
      "url": "https://<tenant>-mcp.<region>.leansignal.io/mcp",
      "headers": { "Authorization": "Bearer <token>" }
    }
  }
}

VS Code — .vscode/mcp.json (workspace) or the user-profile mcp.json; used by Copilot agent mode:

{
  "servers": {
    "leansignal": {
      "type": "http",
      "url": "https://<tenant>-mcp.<region>.leansignal.io/mcp",
      "headers": { "Authorization": "Bearer <token>" }
    }
  }
}

Codex CLI — ~/.codex/config.toml:

[mcp_servers.leansignal]
url = "https://<tenant>-mcp.<region>.leansignal.io/mcp"
http_headers = { "Authorization" = "Bearer <token>" }

Gemini CLI — ~/.gemini/settings.json (global) or .gemini/settings.json (per project):

{
  "mcpServers": {
    "leansignal": {
      "httpUrl": "https://<tenant>-mcp.<region>.leansignal.io/mcp",
      "headers": { "Authorization": "Bearer <token>" }
    }
  }
}

JetBrains — Settings → Tools → AI Assistant → Model Context Protocol → Add, then enter the URL and header (header support varies by IDE version).

What the tools cover

The MCP tools mirror the app’s Stored / Available split: stored tools query central storage (the demanded subset, 30-day retention), while available_* tools query the agent’s local stores — everything collected at the edge, for a short window — and need a connected agent. Control-plane tools list agents, demands, dashboards, and alert rules. As with LeanBuddy, an empty stored result usually means not demanded, not not happening — check the demand set before concluding.

Revoke a token

Revoke a token from the same list on the Preferences page. Revocation deletes it permanently and any client still using it loses access immediately. Revoke tokens you no longer use, and rotate a token immediately if it may have leaked (for example, committed to a repository).