llms.txt and the API
Agents don’t need these docs. They can ask Bunker directly.
| URL | What |
|---|---|
/llms.txt |
A short guide for agents: how to authenticate, deploy, the sandbox rules, the SDK and server functions (api/<name>.js, the server bunker object, secrets, triggers) |
/llms-full.txt |
The same, plus every API operation with its parameters |
/api/v1/agent-guide |
The full agent guide (authentication, deploy, sandbox, SDK, operations) |
/api/v1/ops |
Every operation as JSON: name, method, path, summary, parameters, and whether it’s page-safe or destructive |
/api/v1/openapi.json |
An OpenAPI document generated from the same list |
/api/v1/libraries |
The built-in front-end libraries |
Everything in the UI is in the API
Section titled “Everything in the UI is in the API”Bunker has one rule for its own code: every dashboard action is an operation in a single registry. The REST API, the CLI (bunker api, bunker ops), both MCP servers and the list above are generated from it. There are no hidden dashboard-only endpoints.
curl -s https://bunker.example.com/api/v1/ops | jq '.ops[] | {name, method, path, summary}'{ "name": "sites.deploy", "method": "POST", "path": "/api/v1/sites/{site}/deploy", "summary": "Upload files (zip, folder, single file or JSON) and publish them" }Calling the API
Section titled “Calling the API”- Tokens:
Authorization: Bearer bk_.... No CSRF header needed. - Browser sessions: unsafe requests (POST, PUT, PATCH, DELETE) must send
X-Bunker: 1and come from Bunker’s own origin. - Errors are JSON
{"error": {"code", "message"}}, with plain-English messages that say what to do next. - Rate limits: 1,200 requests a minute per person (burst 300) and 30 a minute per IP for anonymous writes (
BUNKER_RATE_USER_PER_MIN,BUNKER_RATE_PUBLIC_PER_MIN). Over the limit you get429withRetry-After.