Connect an AI assistant (MCP)

TimeTriggers runs a Model Context Protocol server. Connect an AI assistant to it and ask about your triggers in plain words: "why did last night's invoice sync fail?", "which webhooks returned a 5xx today?", "retry the ones that timed out".

URL: https://api.timetriggers.io/mcp
Transport: Streamable HTTP
Authentication: your API key, as Authorization: Bearer YOUR_API_KEY (or ttr-api-keyheader). The assistant sees the key's project only.

Create a key for the assistant in the dashboard, so you can delete it without affecting your other integrations.

Set up your client

Claude Code

claude mcp add --transport http timetriggers https://api.timetriggers.io/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Cursor, VS Code and other clients with a JSON configuration

{
  "mcpServers": {
    "timetriggers": {
      "url": "https://api.timetriggers.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

The key name varies by client (mcpServers, servers…); see your client's documentation.

Clients that only start local servers (stdio) can reach it through mcp-remote:

{
  "mcpServers": {
    "timetriggers": {
      "command": "npx",
      "args": [
        "mcp-remote", "https://api.timetriggers.io/mcp",
        "--header", "Authorization: Bearer YOUR_API_KEY"
      ]
    }
  }
}

The server doesn't support OAuth, so connectors that require it, such as custom connectors on claude.ai, can't use it yet.

Tools

Tool	What it does	Changes anything
`list_triggers`	Lists triggers with their latest outcome. Filters: status, title, custom key, tag.	No
`get_trigger`	One trigger's request and all its attempts.	No
`list_executions`	The request log across triggers. Filters: trigger, custom key, status, HTTP status (`404`, `5xx`…), dead letters only.	No
`get_execution`	One attempt with the response headers and body your target sent.	No
`replay_trigger`	Sends a `completed` trigger's request again, now. See Replay a trigger.	Yes: calls your target
`cancel_trigger`	Cancels a trigger or a recurring trigger, by ID or custom key, like `/cancel`.	Yes

replay_trigger and cancel_trigger are marked as non-read-only, so clients that ask before running such tools ask you first. The read tools use the trigger log endpoints.

What the assistant sees. The read tools return your triggers' URLs, request bodies and your targets' responses. Header values whose name suggests a credential (Authorization, Cookie, Set-Cookie, names containing token, secret, password, api-key or signature) are replaced with [redacted]. Other secrets, for example in a URL or a body, are sent as they are. The REST endpoints return everything unredacted.

Good to know

Request bodies are cut at 4,000 characters in get_trigger. Response bodies are stored up to 10 KB; see Response bodies.
The server keeps no session: every request carries the key, and nothing is pushed to the client. Clients that open a server-to-client stream (GET /mcp) get 405, which the protocol allows.
Tool calls don't use quota.

Connect an AI assistant (MCP)#

Set up your client#

Tools#

Good to know#

Connect an AI assistant (MCP)

Set up your client

Tools

Good to know