Scope the capabilities page and add Cloud availability per row

[hmishra2250]

Stack

These three PRs all edit the same navigation lines in docs.json, so they are stacked in this order:

1. docs: add Browser Sandbox page to navigation #1286, base main. Adds features/browser to the Features nav.
2. This PR (Scope the capabilities page and add Cloud availability per row #1364), base docs/add-browser-to-navigation. Depends on docs: add Browser Sandbox page to navigation #1286.
3. Fix interact sample variables, dead paths, and crawler access guidance #1369, base docs/capabilities-and-licensing. Depends on this PR.

This PR resolves the placement of features/browser by putting it in an "Interact / Browser Sandbox" group together with features/interact, so the page appears exactly once in the nav. Nothing in #1286 needs to change.

Review and merge bottom up. Retarget to main after the parent merges; the diff then shrinks to this PR's own changes.

Base branch note: #1286 comes from a fork, so its head cannot be used as a base directly. docs/add-browser-to-navigation in this repo is #1286's single commit replayed onto current main, byte identical in content, and exists only to carry the stack.

---

Two things to know before reviewing.

Licensing is deliberately excluded from this PR, pending Legal. The page was originally
capabilities-and-licensing.mdx; the ## Licensing section, the licence-dependent misconception
row, and all three [LEGAL REVIEW: confirm] placeholders are removed from the published page and
ship nowhere in this repo. The draft is held outside the repo for Legal and must not be added back
in any form, navigated or not, until Legal signs off — Mintlify has no proven mechanism in this
repo to keep a committed file out of the public site.

The Browser Sandbox navigation addition is included. docs.json nests features/browser
under a new "Interact / Browser Sandbox" nav group and renames the API-reference group to
"Interact / Browser Sandbox Endpoints". If features/browser was a deliberate soft launch, drop
that part of the docs.json change and the rest of this PR still stands.

Summary

Third-party pages assert Firecrawl capability facts that Firecrawl's own snapshots contradict, and
the Interact and Browser Sandbox surfaces were not addressable at all: features/browser was absent
from navigation. This branch adds a canonical capabilities.mdx with a Cloud versus self-hosted
availability column repeated inline in every section, puts Browser Sandbox in the nav, and removes
the licensing section from the published page pending Legal. It also fixes five occurrences of a
broken anchor the original branch was shipping.

Why (evidence)

Source reports: Weekly Deep Insights DI-2026-09-03-WEEKLY (2026-09-03) and DI-2026-08-30-WEEKLY
(2026-08-30); Developer-bank positioning read
(analysis/loop/run/DEVELOPER-POSITIONING-READ.txt, 2026-08-31).

Findings: F5-CORRECTION-QUEUE, thesis "current third-party pages assert Firecrawl capability and
licence facts that other current snapshots contradict". Grade C plus, verdict file
verdicts/deep-F5-corrections.md, where only 2 of 7 targets were double-cited but those two are the
ones the branch acts on. OB-03-interactive-capability-not-addressable, grade B in the
2026-09-03 review: "documented but not addressable: it lives inside extraction pages behind a bounty
and a Preview label." Developer read hypothesis H3: the Interact sandbox is invisible to the
models reasoning about Firecrawl, with 0 of 140 developer answers mentioning Interact or Browser
Sandbox and 4 mentioning the legacy actions array; adjudications D2, D4, D12 and D13 are STALE.

Verified evidence: aimultiple.com asserts "Node.js only (no Python bindings)" against snapshots
showing from firecrawl import FirecrawlApp; dev.to asserts "Apache 2.0 licence" and "103,000
GitHub stars" against a README snapshot showing AGPL-3.0 core with MIT SDKs and 175.7k stars;
knowledgesdk.com asserts "Firecrawl cannot fill out a form." against docs/features/interact.
Developer traces VDEV-202, VDEV-222, VDEV-247 and VDEV-953 reason about actions as if it were the
whole product.

Availability facts are sourced to contributing/open-source-or-cloud.mdx:30 ("Agent, Browser,
managed dashboards, enhanced proxy paths, and enterprise controls are delivered through Cloud") and
the self-hosted feature-support table at contributing/self-host.mdx:175-180.

Final verdict: APPROVE per the resolution addendum to
review-grounding/FINAL-REVIEW-docs-mcp.md. The review body records REQUEST CHANGES; that blocker
is described and resolved below.

Changes

• capabilities-and-licensing.mdx becomes capabilities.mdx, git-tracked as a rename, 192 lines.
  Sections: opening paragraph, ## Capabilities at a glance, 10 capability sections with runnable
  examples, ## Common misconceptions, GitHub-figures note.
• The ## Capabilities at a glance table carries Capability | Availability | Documented in.
  Behavior contract: every capability section repeats its availability inline, so the fact survives
  being retrieved in isolation, which was the original defect. Interact, Browser Sandbox, login and
  session reuse, and screenshots and page actions are marked Cloud only, citing
  /contributing/self-host#self-hosted-feature-support and
  /contributing/open-source-or-cloud#use-firecrawl-cloud-when. JavaScript rendering, PDF parsing,
  and Search are Cloud and self-hosted; structured JSON extraction on self-hosted requires an
  OpenAI-compatible provider or Ollama. Two rows go beyond the brief (screenshots and page actions;
  the provider requirement for structured extraction) because the same self-host table states them
  and their absence would reproduce the same false-composite defect.
• The opening paragraph drops the unconditional SDK-licence sentence and states Search correctly on
  the first screen: "returns titles, descriptions, and URLs, and full page content for every result
  when scrapeOptions is supplied."
• The Search row and section state the default (titles, descriptions, URLs plus query-relevant
  Highlights) and the scrapeOptions / scrape_options contract, per features/search.mdx:32 and
  :34.
• The self-host snippet is now git clone, cd firecrawl, git checkout v2.11.162, with prose
  pointing at the mandatory .env step and the docker compose invocation at
  contributing/self-host.mdx#configure-the-evaluation-deployment. The previous two-liner ran
  docker compose in the parent directory and created no .env.
• The ## Licensing section (8-row table plus AGPL Note), the licence-dependent misconception row,
  and all three [LEGAL REVIEW: confirm] placeholders are removed from the published page. The
  misconceptions table keeps 5 rows: the licence row is replaced by a self-hosting-parity row.
• docs.json: features/browser nested under a new "Interact / Browser Sandbox" nav group; the
  API-reference group renamed from "Interact Endpoints" to "Interact / Browser Sandbox Endpoints";
  the Get Started entry retargeted from capabilities-and-licensing to capabilities.
• features/interact.mdx gains the jobs-to-be-done opener from OB-03.
• Anchor bug fixed: the OB-03 opener and the capability page both linked
  #persistent-profiles-with-scrape--interact with a double dash. The live slug is
  #persistent-profiles-with-scrape-+-interact, verified against
  https://docs.firecrawl.dev/features/interact. Five occurrences corrected. This was shipping
  broken in the original branch.

Blocker resolved in commit 4d8be852. The branch had preserved the licensing content at
drafts/licensing-DRAFT.md on the claim that an unnavigated .md file has no URL. That claim is
false: https://docs.firecrawl.dev/reconciliation/INITIAL_DRIFT_REPORT returns 200 with a full
rendered Mintlify page, a title and an OG image, and so does .../CLAUDE, while
.../reconciliation/THIS_DOES_NOT_EXIST_XYZ returns 404 as a control. Absence from docs.json
removes the sidebar link, not the route, so the branch would have published a live public page whose
first line reads "DRAFT — Licensing (BLOCKED ON LEGAL — DO NOT PUBLISH)". No provably unrouted
in-repo location exists: docs.json has no ignore or exclude key of any kind (its only
top-level keys are $schema, colors, contextual, favicon, footer, integrations, logo, name, navbar, navigation, redirects, seo, theme), and the one live 404 found, docs.firecrawl.dev/README, is
Mintlify special-casing that literal root filename with no second README.md in the tree to confirm
it generalizes. The file was therefore removed with git rm; the draft content is preserved for
Legal in scratchpad/product-prs/D2-capabilities-licensing-v2.md and ships nowhere.

Diffstat: 3 files changed, 210 insertions, 2 deletions.

Verification

docs.json parses as valid JSON.

Nav resolution: 1,772 nav entries, 1,430 unique, 0 missing files. This is the check that catches the
rename regression.

Internal links on the touched files: all ](/…) targets in capabilities.mdx and
features/interact.mdx exist as .mdx files, 0 broken.

Anchors verified against the live rendered page rather than a local slug guess, which is how the
double-dash bug surfaced: #persistent-profiles-with-scrape-+-interact confirmed present in the
served HTML from curl https://docs.firecrawl.dev/features/interact, plus
#use-firecrawl-cloud-when (:27), #self-hosted-feature-support (:170),
#configure-the-evaluation-deployment (:73), #interact-via-prompting (:105),
#running-code (:126), #session-lifecycle (:228), and /features/search-highlights. HTTP 200
on all pages checked.

Negative control on published content: grep across all *.mdx excluding drafts/ returns zero hits
for LEGAL REVIEW and zero for SDKs are MIT. The only AGPL and MIT strings in published .mdx
are pre-existing ones in v1/introduction.mdx, its localizations, and
learn/guide/firecrawl-ui-template.mdx, none touched by this branch. No dangling reference to
capabilities-and-licensing remains in any non-localized .mdx or .json.

Diff audit for stray unnavigated content: git diff --name-status origin/main...HEAD lists only
capabilities.mdx (new, navigated), docs.json, features/interact.mdx and the removed
drafts/licensing-DRAFT.md.

mint broken-links under Node 22.23.2: 208 broken links in 95 files, identical to baseline; neither
capabilities.mdx nor features/interact.mdx appears. mint validate: the same 17 pre-existing
locale snippet warnings; no touched file appears.

Localized trees untouched, per the repo CLAUDE.md.

Not in this PR

• Legal sign-off on the licensing section. The full draft, including the licence table, the
  AGPL-obligations placeholder, the removed misconception row and the "what must not be said" list,
  lives only in scratchpad/product-prs/D2-capabilities-licensing-v2.md. It must not be added back
  to the docs repo in any form, navigated or not, until Legal signs off, because Mintlify has no
  proven mechanism in this repo to keep a committed file out of the public site. Two things block
  publication: Legal must confirm the SDK licence position, and Core must fix the underlying
  declarations.
• A Core PR, drafted but not filed. Core was not modified. Title:
  chore(licensing): fix stale GPLv3 declaration and add missing SDK LICENSE files. Item 1:
  apps/python-sdk/setup.py:67 declares license="GNU General Public License v3 (GPLv3)",
  contradicting apps/python-sdk/LICENSE (MIT) and apps/python-sdk/pyproject.toml:22
  (license = {text = "MIT License"}); that string is what lands in PyPI package metadata for any
  build still going through setup.py. Item 2: apps/go-sdk, apps/rust-sdk, apps/php-sdk,
  apps/java-sdk and apps/dot-net-sdk ship no LICENSE file, while README.md:905 describes the
  SDKs collectively as MIT, which is not a per-package grant under an AGPL-3.0 repository.
• An Agent row in the capabilities table. Agent is covered in the intro paragraph and the
  self-hosting row but has no row of its own, so its Cloud-only status does not survive isolated
  retrieval the way every other capability's does. Optional nit from the review.

Links

Files changed:

• capabilities.mdx (renamed from capabilities-and-licensing.mdx)
• docs.json
• features/interact.mdx

Evidence packet:
agent-experience-deepinsights-cleanroom/artifacts/deep-insights-sep3-verification-20260904/

Licensing draft for Legal:
scratchpad/product-prs/D2-capabilities-licensing-v2.md, section "Licensing draft (for Legal; not
in repo)".

---

Added 2026-09-06

Monitor is in the capabilities table. The page is the canonical answer to what Firecrawl can do, and the table listed SDKs, Interact, Browser Sandbox, JavaScript rendering, Parse, Search, structured extraction, screenshots and self-hosting. It omitted Monitor, which ships eight public v2 routes, eight MCP tools, a CLI verb and four docs pages in five languages. Deep Insights DI-2026-09-04-WEEKLY, finding OB-04-MONITORING-CATEGORY-FILING: 27 of 28 monitoring answers name a competitor and never mention Firecrawl. One row is added, in the table's existing format, pointing at Monitoring.

Verification, rerun on this branch rebased onto origin/main at c754d6b6:

• mintlify broken-links (Mintlify CLI 4.2.773, Node 22.23.2): 208 broken links, byte-identical to the same run on a clean origin/main worktree. None in capabilities.mdx.
• mintlify validate: 17 warnings, all "Could not find file /snippets/...", the same set as on origin/main.

Note for the merge order

Resolved 2026-09-07 by stacking. See the Stack section at the top. This PR and #1286 both edit the Features group in docs.json at the same place. origin/main docs.json line 195 is "features/interact". #1286 inserts "features/browser", on the line above it. This PR replaces that line with an "Interact / Browser Sandbox" group holding both pages, which is the intent-preserving resolution: features/browser appears exactly once, inside the group. This branch is now rebased on top of #1286's change with that resolution applied, so there is no conflict left to settle at merge time.

Added 2026-09-07

The capabilities table now answers five questions per row instead of one. The table had a single Availability column, so it said where a capability runs and nothing else. A reader could not tell from it whether a capability has an MCP tool, how long a browser session lasts, whether the endpoint takes a request with no API key, or whether they need credentials for the site they are reading. Those answers do not track deployment: Browser Sandbox is Cloud-only and has no MCP tool, Interact is Cloud-only and accepts a keyless request, and persistent profiles need your login for the target site while nothing else in the table does.

Availability is replaced by Deployment, MCP tool, Session lifetime, API auth and Target credentials, and a short list above the table says what each column means. Reference: Deep Insights DI-2026-09-05-WEEKLY, Astra edition, finding OB-04.

Every cell is read off origin/main in the repo that owns the fact.

• MCP tool names: firecrawl-mcp-server src/index.ts for firecrawl_scrape, firecrawl_search, firecrawl_parse, firecrawl_interact, firecrawl_interact_stop and firecrawl_agent, and src/monitor.ts for the eight firecrawl_monitor_* tools.
• Session lifetime: apps/api/src/controllers/v2/browser.ts:44-45 and apps/api/src/controllers/v2/scrape-browser.ts:74-75 give ttl 30 to 3600 seconds with a default of 600, and idle activityTtl 10 to 3600 seconds with a default of 300. The scrape request timeout and its per-format increases are apps/api/src/controllers/v2/types.ts:797 and :871-897.
• API auth: the allowKeyless option on each route in apps/api/src/routes/v2.ts, with the middleware default at apps/api/src/routes/shared.ts:245-253, where an omitted option means a key is required. /v2/scrape, /v2/search, /v2/parse and /v2/scrape/{id}/interact accept a keyless request. /v2/browser, /v2/extract, /v2/agent and the /v2/monitor routes require a key.
• Target credentials: the profile field at apps/api/src/controllers/v2/browser.ts:50-54 and apps/api/src/controllers/v2/scrape-browser.ts:79-82.

Two cells say a capability has no MCP tool rather than filling the cell in. The standalone Browser Sandbox has none, because the MCP interact tools act on an existing scrape and no browser tool is registered anywhere in src. Self-hosting has none because it is a deployment, not an endpoint. The column header points at MCP tools, which already documents that the working tool set depends on the API the server is connected to, so the column gives the tool name rather than a per-deployment claim.

Verification on this branch, Node 22.23.2:

• mintlify broken-links: 208 broken-link entries across 95 files, identical to the same run on a detached worktree at this branch's previous head 26ed2f0f. None in capabilities.mdx.
• mintlify validate: 17 warnings, the same count as the detached baseline.
• git diff --name-only origin/docs/add-browser-to-navigation...HEAD returns capabilities.mdx, docs.json and features/interact.mdx only.

Added 2026-09-07

Qualified the Monitor row's deployment cell in capabilities.mdx after external review. It previously read "Cloud and self-hosted (core route)" with no condition. Self-hosted scheduled operation is gated. On firecrawl origin/main, apps/api/src/services/queue-worker.ts:467 starts the monitor scheduler and both check consumers only behind if (config.USE_DB_AUTHENTICATION && !config.DISABLE_MONITORING) {, and the else branches log "Skipping monitor worker startup because database authentication is disabled" (:490) and "Skipping monitor worker startup because NUQ_RABBITMQ_URL is not configured" (:494). DISABLE_MONITORING defaults to false at apps/api/src/config.ts:354, but the default compose file turns database authentication off, at docker-compose.yaml:33: USE_DB_AUTHENTICATION: ${USE_DB_AUTHENTICATION:-false}. A default self-hosted stack therefore has the core implementation and never runs a scheduled check. The cell now says the core implementation is available self-hosted but scheduled operation requires database authentication and the monitoring dependencies to be configured, and links Self-host Firecrawl, which documents USE_DB_AUTHENTICATION. The nav block and the licensing content are untouched. mintlify broken-links on Node 22.23.2 is identical to a detached worktree at the previous head 93dc8d5e.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
firecrawl	🟢 Ready	View Preview	Sep 4, 2026, 4:56 PM

💡 Tip: Enable Automations to automatically generate PRs for you.