serving-the-api

Serving the API

kreuzcrawl serve starts a long-running REST API server exposing the crawl

engine over HTTP. The endpoints are Firecrawl v1-compatible, so existing

Firecrawl clients and SDKs can point at a local kreuzcrawl instance.

This subcommand requires the binary to be built with the optional api

feature (non-default). The Homebrew tap is built with all features, so serve

is available there; a from-source build needs

cargo install --git https://github.com/kreuzberg-dev/kreuzcrawl kreuzcrawl-cli --features api

(or --features all) to include it.

Quick recipe

kreuzcrawl serve --host 127.0.0.1 --port 3000

It logs Starting REST API server on <host>:<port> to stderr and blocks until

terminated. Hit GET /health to confirm it is up.

Flags

| Flag | Default | Purpose |

| -------- | --------- | -------------------------------- |

| --host | 0.0.0.0 | Address to bind to. |

| --port | 3000 | Port to listen on. |

serve does not take the per-request crawl flags — it boots with the default

CrawlConfig and accepts per-request options in each HTTP request body.

Endpoints

Firecrawl v1 surface:

| Method + path | Purpose |

| ------------------------- | ---------------------------------------------------- |

| POST /v1/scrape | Scrape a single URL (synchronous). |

| POST /v1/crawl | Start an asynchronous crawl job → returns a job id. |

| GET /v1/crawl/{id} | Poll crawl job status and collect results. |

| DELETE /v1/crawl/{id} | Cancel a running crawl job. |

| POST /v1/map | Discover URLs on a site (synchronous). |

| POST /v1/batch/scrape | Start an asynchronous batch-scrape job → job id. |

| GET /v1/batch/scrape/{id} | Poll batch-scrape job status and results. |

| POST /v1/download | Download a document from a URL. |

Operational:

| Method + path | Purpose |

| ----------------- | ---------------------------------------- |

| GET /health | Health check ({ status, version }). |

| GET /version | Version information. |

| GET /openapi.json | OpenAPI schema for the whole surface. |

Request bodies use camelCase fields. Scrape/crawl/map mirror the CLI:

url, formats, onlyMainContent, includeTags/excludeTags, timeout

for scrape; url, maxDepth, maxPages, includePaths/excludePaths for

crawl; url, limit, search for map. Responses wrap data as

{ success, data | error }; async jobs return { success, id } and are polled

for { status, total, completed, data, error }.

Examples

Scrape via HTTP

curl -s http://127.0.0.1:3000/v1/scrape \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com","formats":["markdown"]}' | jq '.data'

Start a crawl job and poll it

id=$(curl -s http://127.0.0.1:3000/v1/crawl \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com","maxDepth":2,"maxPages":100}' | jq -r '.id')

curl -s "http://127.0.0.1:3000/v1/crawl/$id" | jq '{status,completed,total}'

Map URLs

curl -s http://127.0.0.1:3000/v1/map \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com","limit":500,"search":"docs"}' | jq '.data'

Operational notes

• The server accepts CORS from any origin and ships **no built-in

authentication** — do not expose it on a public interface. Bind to

127.0.0.1 for local use, or put it behind a reverse proxy / gateway that

enforces auth and rate limiting before exposing it more widely.

• Request bodies are capped at 10 MB and each request times out after 5

minutes.

• Crawl and batch-scrape are asynchronous: the POST returns a job id

immediately and you poll the {id} endpoint until status is completed,

failed, or cancelled.

When to run serve vs. CLI vs. MCP

• CLI (scrape/crawl/map/interact) — one-shot jobs, shell pipelines,

scripting. No long-running process.

• MCP (kreuzcrawl mcp) — driving the crawler from an AI agent harness

(Claude Code, Codex, Cursor, Gemini, opencode). Auto-registered by this

plugin; prefer it inside an agent.

• serve (kreuzcrawl serve) — a persistent HTTP service for other

applications or non-agent clients that speak REST, or when you want

Firecrawl-v1 compatibility. Choose it when many callers need the crawler

over the network rather than one agent or one shell.