Firecrawl crawl credits per page, and how Sume meters crawl_site

Firecrawl's crawl page says each crawled page costs 1 credit and that JSON mode adds 4 credits per page. Sume's crawl_site tool is listed in the MCP docs as a write tool and an "unbilled utility", and its description calls it a "Bounded unbilled utility rollout." Sume states no per-page credit rate for it.

Firecrawl facts are from its crawl page; Sume facts from MCP tools and gates and the tool description in the repo, all read 2026-09-30.

How does Firecrawl count credits for a crawl?

The page states the rate per page: 1 credit, and 4 additional credits per page when JSON mode is used. A status poll of GET /v2/crawl/{id} returns status, completed, total and creditsUsed, so you can read spend while the crawl runs. For a crawl with JSON extraction on, the per-page figure from the page is therefore 1 + 4 credits.

How is crawl_site labelled on Sume?

Does JSON mode change anything on Sume?

The tool description says crawl_site accepts the same formats and JSON options as crawl_scrape for each page, plus only_main_content, timeout_ms, wait_for_ms, actions and fresh. The docs do not attach a price to turning JSON options on. Treat "unbilled" as the current rollout label, not a promise about future pricing, and check the docs again before budgeting a large job.

What are the other bounds on a Sume crawl?

The description gives limit as at most 20 pages (default 10) and requires a stable idempotency_key. Start it once, then call jobs_wait on the same id and read pages with crawl_get; do not start a second crawl to poll. For waiting on long jobs see MCP jobs_wait.

What should I do when budgeting a crawl?

If you run Firecrawl directly, multiply pages by 1 (or 5 with JSON mode) and watch creditsUsed. If you call crawl_site through Sume's MCP server, plan for the 20-page bound, keep the job_id, and re-read the MCP docs for any change to the unbilled label.