Skip to main content
Tallwatch hosts the MCP server. There is no npm package, no npx, and no local process (Claude Desktop is the exception below). Point the client at https://api.tallwatch.com/mcp with a Bearer . The agent can check org status, list monitors, triage incidents, and (with a write key) pause or ack. Same surface as the for ops work. The app UI still uses your session cookie.

Setup

  1. Create a key in . The tw_live_… secret is shown once.
  2. Copy prompt on that page pastes setup into the agent, or pick your client below.
  3. Keep the key in user-level config, not a committed project file.
Each client has its own config shape. The wrong URL field or transport type is ignored, or it falls back to SSE and never reaches /mcp.
User: ~/.cursor/mcp.json. Project: .cursor/mcp.json. Remote entries use url and headers. Do not set type.
You can put the secret in the environment instead: "Authorization": "Bearer ${env:TALLWATCH_API_KEY}". Reload MCP in Cursor Settings.

Tools

Same ops surface as the (minus browser login / browse). Prefer get_status when the user asks what is broken. UptimeRobot import is in the app (Monitors → Import) and CLI — not an MCP tool.

Create / update monitors

create_monitor and update_monitor use the shared CreateMonitor contract:
  • Non-heartbeat: at least three regions.
  • Fields: name, type, config, regions, interval_sec (optional notify_after_sec, escalation_policy_id, project_id).
  • Consensus is fixed (m=1 / k=2). Do not send consensus.
  • Removed fields (depends_on_monitor_ids, tags) are rejected / ignored — do not send them.
MCP uses the same key as the REST API, so apply. Prefer one list_monitors over many get_monitor calls. . . .