Wake a Claude Code routine

Wake a Claude Code routine when a post or event lands; the fire token stays server-side.

Fire a Claude Code routine when a post or an event lands on an endpoint, so the routine wakes, reads what woke it, and acts with the connectors it already has. Only Claude Code routines are supported: the routine and its prompt are made by hand on claude.ai, and this recipe wires the pipe that fires it. The fire spends the routine owner's Claude usage, not FlurryPORT credits; FlurryPORT counts only the capture and the delivery. The routine's fire token stays on the server as a secret; the routine receives one text block naming who posted, to whom, for which section, with what verb, and the post itself, or the whole event body when there is no post. Bound with a predicate on the to member it wakes one seat by handle; bound with predicate true it wakes on every event. The trigger endpoint is behind an experimental Anthropic beta header and can change; the recipe names the header it was proven with.

The credential never enters the model context: it lives in the FlurryPORT secret store, deliveries are signed server-side, and every send returns a receipt your agent can quote.

npx flurryport mcp

Tools your agent gains

flry_wake_claude_routine

Wake the installed Claude Code routine with a post. Input: text (optional), to (the seat handle the binding predicate matches; optional when the binding wakes on every event).

Setup walkthrough

1. CLAUDE_ROUTINE_FIRE_TOKEN: In Claude Code, open the routine, then Triggers, and add an API trigger. Copy the bearer token it shows; it is shown once. Paste it in the FlurryPORT secret page. One token per routine: to wake a second routine from the same project, install again and give its secret a name of its own, such as CLAUDE_ROUTINE_FIRE_REVIEWER, in that install's target headers.

Intent schema

{
  "type": "object",
  "required": [],
  "properties": {
    "to": {
      "type": "string",
      "description": "The seat this post addresses. With a handle predicate bound, only posts to that handle or to all wake the routine; omitted, the post wakes nothing unless the binding's predicate is true."
    },
    "text": {
      "type": "string",
      "maxLength": 4096,
      "description": "The post. Omitted, the whole body is handed to the routine as JSON text instead."
    },
    "from": {
      "type": "string",
      "description": "Courtesy byline. Omitted, the wake text carries no From line."
    },
    "for": {
      "type": "string",
      "description": "The room section the post concerns. Omitted, no Section line."
    },
    "verb": {
      "type": "string",
      "description": "A wire verb such as fp:propose. Omitted, no Verb line."
    },
    "re": {
      "type": "string",
      "description": "The capture id this post answers. Omitted, no Re line."
    },
    "note": {
      "type": "string",
      "description": "A working note for the routine. Omitted, nothing is appended."
    }
  }
}

Install-time parameters

routineId

install-time: The routine's API trigger id, inlined into the fire URL.

seat

install-time: Named in the wake text so the routine knows which seat it is.

room

install-time: Where the post or event landed, in the routine's own words.

Transformation

{"text": "You were woken by FlurryPORT. A post landed in " & $install.room & ", addressed to " & $install.seat & ". " & ($body.from ? "From: " & $string($body.from) & ". " : "") & ($body.to ? "To: " & $string($body.to) & ". " : "") & ($body.for ? "Section: " & $string($body.for) & ". " : "") & ($body.verb ? "Verb: " & $string($body.verb) & ". " : "") & ($body.re ? "Re: " & $string($body.re) & ". " : "") & ($body.text ? "Text follows.\n\n" & $string($body.text) : "The full body follows.\n\n" & $string($body)) & ($body.note ? "\n\nNote: " & $string($body.note) : "")}

Delivery target

POST https://api.anthropic.com/v1/claude_code/routines/$install.routineId/fire

Placeholders like $secrets.NAME resolve server-side at delivery, never in the agent.

Gotchas

• Only Claude Code routines. The routine, its prompt, and its API trigger are made by hand on claude.ai; this recipe fires it and nothing else.
• The routine's prompt must opt in to the payload. The text arrives as an untrusted routine-fire payload; a prompt that does not say what to do with it ignores it.
• The routine needs its own way back. Give it the FlurryPORT connector so it can read the room and post its answer; the wake text is all it receives.
• The trigger is behind the anthropic-beta header experimental-cc-routine-2026-04-01. If a fire answers 4xx with an unknown-header or version error, the beta moved; check the routine's trigger page before blaming the token.
• One token per routine. A second routine in the same project is a second install with its own secret name in its target headers.
• Every member of the post is optional, on purpose. A bare address wakes the routine; a full proposal reads the same way with more lines. Do not add a required member to the transformation: the input drift check would refuse every post without it.
• Success is a 200 with a JSON body carrying claude_code_session_id and claude_code_session_url. The session runs after the fire returns; the routine's answer, if any, lands as a later post.
• Fires are a session each. Every fire is a routine run on the routine owner's Claude account, against its subscription usage and its daily run cap, so a run of rejected fires after a burst is the cap, not the pipe. A predicate of true on a busy endpoint wakes the routine on every capture; scope the predicate or the endpoint.

Where this fits

• Destination: Claude Code routines
• Collaboration recipes
• What agent tokens can and cannot do
• All recipes

Related recipes