Changelog

This page is generated from the root CHANGELOG.md, which is maintained by release-please during releases.

:::note The source of truth is the repository root changelog. Do not edit this docs page manually. :::

All notable changes to CRW are documented here.

0.3.4 (2026-04-09)

Bug Fixes

  • config: env var overrides not applied due to missing prefix_separator (71c7ae5), closes #18

0.3.3 (2026-04-09)

Features

  • add APT/Debian package distribution (c34b8e9)
  • renderer: spawn all available browsers for multi-renderer fallback (f546437)

0.3.2 (2026-04-08)

Bug Fixes

  • cli: auto-prepend https:// when no scheme provided (1050606)

0.3.1 (2026-04-08)

Features

  • add llms.txt, SKILL.md, MCP init command, and docs UI improvements (1b22d19)
  • add one-line install script with auto platform detection (6354f79)
  • docs: add dark mode logo support and improve docs UI (047df7b)
  • docs: align design with SaaS site and update branding (631d07c)
  • docs: unify docs into docs.fastcrw.com with Mintlify-style design (4994998)
  • docs: update URLs, dark mode, syntax highlighting, and benchmarks (0678cdf)
  • release all 3 binaries, CLI auto-browser, README overhaul (aa2950d)
  • update README banner with new logo (bcba1ad)

Bug Fixes

  • crawl HTTP polling bug + SDK test suite + docs (#16) (b6d8983)
  • remove internal implementation detail from roadmap (a5013f0)

0.3.0 (2026-04-02)

Features

  • add search() method to Python SDK and docs (591e3fe)

0.2.2 (2026-04-02)

Bug Fixes

  • renderer: escalate to JS renderer on HTTP 401/403 responses (f515caa)
  • use GitHub latest release instead of pinned version for binary download (4afcb1a)

0.2.1 (2026-03-28)

Bug Fixes

  • make crw-mcp npm wrapper executable (576a9eb)
  • use latest tag in server.json OCI identifier (7ec3b82)

0.2.0 (2026-03-28)

Features

  • add MCP Registry support for official server discovery (154b9f5)

0.1.2 (2026-03-27)

Bug Fixes

  • vendor pdf-inspector as crw-pdf for crates.io publishability (3f7681d)

0.1.1 (2026-03-26)

Bug Fixes

  • skip already-published crates without masking real errors (010649c)

0.1.0 (2026-03-26)

Features

  • add PDF extraction support via pdf-inspector (06dd5bf)

0.0.14 (2026-03-25)

Features

  • mcp: auto-download LightPanda binary for zero-config JS rendering (41f443b)
  • mcp: auto-spawn headless Chrome for JS rendering in embedded mode (9a6b0ae)

Bug Fixes

  • ci: move crw-mcp to Tier 4 in release workflow and add workflow_dispatch (d7584a8)

0.0.13 (2026-03-24)

Features

  • mcp: add embedded mode — self-contained MCP server, no crw-server needed (75e5450)

Bug Fixes

  • ci: switch release-please to simple type for Rust workspace support (51cd420)

v0.0.12

  • Readability drill-down — when <main> or <article> wraps >90% of body, the extractor now searches inside for narrower content elements (.main-page-content, .article-content, .entry-content, etc.) instead of discarding. Fixes MDN pages returning 35 chars and StackOverflow returning only the question
  • Base64 image stripping — data: URI images are removed in both HTML cleaning (lol_html) and markdown post-processing (regex safety net). Eliminates massive base64 blobs from Reddit and similar sites
  • Select/dropdown removal — <select> elements removed in onlyMainContent mode; dropdown/city-selector/location-selector noise patterns added. Fixes Hürriyet city dropdown leaking into content
  • Extended scored selectors — added .main-page-content, .js-post-body, .s-prose, #question, .page-content, #page-content, [role="article"] for better MDN, StackOverflow, and generic site coverage
  • Smarter fallback chain — when primary extraction produces too-short markdown, both fallbacks (cleaned HTML and basic clean) are tried and the longer result is picked, instead of short-circuiting on non-empty but insufficient content

v0.0.11

  • Stealth anti-bot bypass — automatic stealth JS injection via Page.addScriptToEvaluateOnNewDocument before every CDP navigation. Spoofs navigator.webdriver, Chrome runtime object, plugins array, languages, permissions API, iframe contentWindow, and toString() proxy to bypass Cloudflare, PerimeterX, and other bot detection systems
  • Cloudflare challenge auto-retry — detects Cloudflare JS challenge pages ("Just a moment", cf-browser-verification, challenge-platform) after page load and polls up to 3 times at 3-second intervals for non-interactive challenges to auto-resolve
  • HTTP → CDP auto-escalation — FallbackRenderer::fetch() in auto mode now checks HTTP responses for anti-bot challenge signatures and automatically escalates to JS rendering when detected, instead of returning the challenge HTML
  • Chrome failover in Docker — full automatic failover chain: HTTP → LightPanda → Chrome. Added chromedp/headless-shell as a Docker Compose sidecar service with 2GB shared memory. If LightPanda crashes on complex SPAs (React, Angular), Chrome handles the render
  • Chrome WS URL auto-discovery — CDP renderer resolves Chrome DevTools WebSocket URL via the /json/version HTTP endpoint with Host: localhost header (required for chromedp/headless-shell's socat proxy). Uses OnceCell for lazy one-time resolution
  • Proxy configuration docs — expanded proxy config comments with examples for HTTP, SOCKS5, and residential proxy providers (IPRoyal, Oxylabs, Smartproxy)
  • Raw string delimiter fix — fixed markdown.rs test that used r#"..."# with a string containing "#, changed to r##"..."##

v0.0.10 / v0.0.9

  • Crawl cancel endpoint — DELETE /v1/crawl/{id} cancels a running crawl job via AbortHandle and returns { success: true }
  • API rate limiting — token-bucket rate limiter (configurable rate_limit_rps, default 10). Returns 429 with error_code: "rate_limited" when exceeded
  • Machine-readable error codes — all error responses now include an error_code field (e.g. "invalid_url", "http_error", "rate_limited", "not_found")
  • Map response envelope — /v1/map now returns { success, data: { links } } instead of { success, links } for consistency with other endpoints
  • Fenced code blocks — indented code blocks (4-space) are post-processed into fenced (```) blocks for better LLM/RAG compatibility
  • Sphinx footer cleanup — "footer" added to exact-token noise patterns, catching <div class="footer"> in Sphinx/documentation sites
  • renderedWith: "http" — HTTP-only fetches now report rendered_with: "http" in metadata instead of null
  • 405 JSON responses — all routes now have .fallback(method_not_allowed) returning structured JSON with error_code: "method_not_allowed" instead of empty bodies
  • Anchor link cleanup — empty anchor links ([](#id), [¶](#id)) and pilcrow/section signs stripped from Markdown output
  • role="contentinfo" cleanup — elements with ARIA roles contentinfo, navigation, banner, complementary removed during cleaning
  • Tiny chunk merging — topic chunking merges heading-only chunks (<50 chars) with the next chunk to improve RAG embedding quality

v0.0.8

  • Wikipedia / MediaWiki onlyMainContent fix — onlyMainContent: true now correctly extracts article text from Wikipedia pages (~49% size reduction). Previously the <html> element's class="vector-toc-available" matched the "toc" noise pattern via substring, removing the entire page
  • 3-tier noise pattern matching — noise class/id matching now uses substring (long patterns), exact-token (short/ambiguous: toc, share, social, comment, related), and prefix (ad-, ads-) matching to avoid false positives
  • Structural element guard — noise handler never removes <html>, <head>, <body>, or <main> elements
  • Re-clean after readability — readability output is re-cleaned to strip residual noise (infobox, navbox, catlinks) that survives inside broad containers
  • Wikipedia-aware readability — added .mw-parser-output, #mw-content-text, #bodyContent to scored selectors; priority/scored selectors that wrap >90% of body are skipped
  • BYOK LLM extraction — per-request llmApiKey, llmProvider, llmModel fields for bring-your-own-key structured extraction without server config
  • JSON format validation — formats: ["json"] without jsonSchema now returns a 400 error instead of a warning
  • Block detection skip — pages >50 KB skip interstitial/block detection (no more false "blocked by anti-bot" on Wikipedia)
  • Null byte URL rejection — URLs with %00 or null bytes rejected at validation
  • Request timeout — default timeout bumped from 60s to 120s
  • Dockerfile fix — corrected cargo build flags, added config.docker.toml

v0.0.7

  • success: false on 4xx targets — scraping a 403/404/429 target with minimal body now correctly returns success: false with error details, instead of success: true with a warning. Targets with real content (custom error pages) still return success: true with a warning
  • JS renderer fallback warning — when renderJs: true is requested but no CDP renderer is available, the response now includes rendered_with: "http_only_fallback" and a warning instead of silently falling back
  • CDP health check — is_available() now runs a real Browser.getVersion command instead of just testing the WebSocket connection
  • Specific error messages — unknown formats now return descriptive errors (e.g., "Unknown format 'extract'. Valid formats: ...") instead of generic 422
  • "extract" format alias — formats: ["extract"] and formats: ["llm-extract"] are now accepted as aliases for "json" (Firecrawl compatibility)
  • Chunk dedup by default — deduplication is now enabled by default for all chunking strategies; separator-only chunks (---, ***) are filtered out
  • Chunk relevance scores — chunks now return { content, score, index } objects instead of plain strings when a query is provided
  • Map timeout — /v1/map accepts a timeout parameter (default 120s, max 300s) to prevent 502s on large sites
  • Stealth + JS rendering fix — stealth: true with renderJs: true no longer bypasses CDP; the shared renderer is used with stealth headers injected
  • BM25 NaN guard — prevents NaN scores when all chunks are empty

v0.0.6

  • Crate READMEs on crates.io — all 7 crates now have detailed README documentation visible on their crates.io pages, with usage examples, API docs, and installation instructions

v0.0.5

  • crw-cli now on crates.io — install the standalone CLI with cargo install crw-cli and scrape URLs without running a server
  • Parallelized release workflow — crate publishing uses tiered parallelism, cutting release time by ~2.25 minutes
  • CLI and MCP install docs — README now includes cargo install instructions for both crw-cli and crw-mcp

v0.0.4

  • Hardened rendering and warning semantics — improved reliability of the rendering pipeline and warning detection logic
  • XPath output escaping — XPath extraction results are now properly escaped to prevent injection
  • Broadened status warnings — expanded HTTP status code range that triggers warning metadata
  • Capped interstitial scan — bounded interstitial page detection to avoid excessive scanning
  • Clippy cleanup — simplified status code checks for cleaner, idiomatic Rust

v0.0.3

  • Warning-aware target handling — 4xx and anti-bot targets now return success: true with warning and metadata.statusCode
  • More reliable JS rendering — CDP navigation now waits for real page lifecycle completion before applying waitFor
  • Stealth decompression fix — gzip and brotli responses decode cleanly instead of leaking garbled binary payloads
  • Crawl compatibility — limit, maxPages, and max_pages now normalize to the same crawl cap
  • XPath and chunking fixes — XPath returns all matches, chunk overlap/dedupe is supported, and scorer rank order is preserved

v0.0.2

  • CSS selector & XPath — target specific DOM elements before Markdown conversion (cssSelector, xpath)
  • Chunking strategies — split content into topic, sentence, or regex-delimited chunks for RAG pipelines (chunkStrategy)
  • BM25 & cosine filtering — rank chunks by relevance to a query and return top-K results (filterMode, topK)
  • Better Markdown — switched to htmd (Turndown.js port): tables, code block languages, nested lists all render correctly
  • Stealth mode — rotate User-Agent from a built-in Chrome/Firefox/Safari pool and inject 12 browser-like headers (stealth: true)
  • Per-request proxy — override the global proxy on a per-request basis (proxy: "http://...")
  • Rate limit jitter — randomized delay between requests to avoid uniform traffic fingerprinting
  • crw-server setup — one-command JS rendering setup: downloads LightPanda, creates config.local.toml

v0.0.1

  • Firecrawl-compatible REST API — /v1/scrape, /v1/crawl, /v1/map with identical request/response format
  • 6 output formats — markdown, HTML, cleaned HTML, raw HTML, plain text, links, structured JSON
  • LLM structured extraction — JSON schema in, validated structured data out (Anthropic tool_use + OpenAI function calling)
  • JS rendering — auto-detect SPAs via heuristics, render via LightPanda, Playwright, or Chrome (CDP)
  • BFS crawler — async crawl with rate limiting, robots.txt, sitemap support, concurrent jobs
  • MCP server — built-in stdio + HTTP transport for Claude Code and Claude Desktop
  • SSRF protection — private IPs, cloud metadata, IPv6, dangerous URI filtering
  • Docker ready — multi-stage build with LightPanda sidecar