What our run established
Our October 3, 2026 run at commit 3c32041 recorded successful installation, build and test commands in our sandbox. The harness records each command's status, exit code, elapsed time and timeout state separately from its parsed test summaries. Installation took 28 seconds, the build took 235 seconds, and the test command took 170 seconds.
The stored parser summaries report 1,602 passed and 0 failed. The attached record warns that repeated Cargo summaries may double-count suites, so these figures cannot establish a unique test total or exhaustive coverage. The raw fetch doctest footer includes two ignored tests. Those tests were skipped, not failed.
Successful commands establish that this checkout built and its test command returned zero in the recorded environment. They do not establish extraction accuracy, successful browser rendering, bot-challenge handling, crawl throughput, MCP client compatibility or hosted API performance. No new execution accompanies this correction. No audit result was recorded, and dependency scan coverage is unknown.
The MrKeyoor testing methodology explains how command outcomes and parsed counts reach reviews. Its supplied November 1 snapshot provides provenance context; it does not prove that the October run used every detail of that later procedure.
Which extraction features work locally?
The CLI and MCP server work locally without an account for core extraction. The README at the recorded commit lists Markdown, compact LLM text, plain text, JSON and cleaned HTML outputs. Structured extraction and summaries need a local or configured language model. Search and research are listed as hosted API tools.
The extraction core has no network I/O and can be used independently of the fetching layer. That separation is useful if your application already fetches HTML and needs to convert it. For an agent workflow, the MCP server exposes the surrounding extraction tools, but our sandbox record does not include a client initialization, tool-listing or scrape session.
Does JavaScript require the hosted API?
The current README documents local Chrome rendering with the browser feature. That differs from the recorded commit's README, which directs users to the hosted API for JavaScript rendering and protected-site access. Read the documentation for the version you intend to deploy rather than treating either description as universal.
Current browser documentation also describes actions, scrolling and persistent profiles, with qualifications around proxies and connected browsers. These are upstream descriptions, not workloads validated by our run. Before choosing a hosted service solely for rendering, check whether your selected Webclaw version provides the local browser path you need and test it on your application pages.
Can EmptyPdf identify every scanned document?
The maintainer says EmptyPdf means zero non-whitespace characters after normalization. In issue 104, the maintainer explains that a scanned page with a page number, footer stamp or partial OCR layer can return text without reaching that error path.
The issue is marked completed, so its original request does not establish that the byte handoff remains missing. The supplied extract also does not establish the completed API's exact behavior at our recorded commit. Test text-free PDFs and scans with incidental text separately. Decide whether OCR is needed from the content your workflow requires, rather than from the presence or absence of this error alone.
Which alternatives fit different workflows?
Firecrawl documents search, scraping and page interaction, with open-source and hosted options. Evaluate it as a replacement when those API operations are central to your application. Its README distinguishes cloud features from self-hosting; this review supplies no comparative success-rate or latency measurement.
Trafilatura is a Python package and CLI for extracting web text, metadata and comments. Evaluate it as a replacement when those outputs are the main requirement and Python fits your application. Its documented discovery and crawling components mean it should not be described as only an article-text function.
An OCR processor would complement these extraction tools when documents need recognition. It would not replace URL discovery or fetching.
What the scores mean
Setup ease is 4 because our source build completed and several installation routes are documented, although native prerequisites add friction. Documentation is 4 because installation and capability boundaries are described, but readers must match browser guidance to their chosen version. Community is unassessed because the supplied evidence does not establish contributor breadth or sustained support reliability. Maturity is 3 because successful commands support a pilot, while workload behavior and unique test totals remain unverified.
These editorial scores use this rubric: 1 means a demonstrated major obstacle; 2 means substantial friction; 3 means usable with material qualifications; 4 means strong support with bounded gaps; 5 means comprehensive evidence within the assessed scope. Unassessed means the evidence cannot support a numerical judgment.
What to test before deployment
Our run did not test a representative extraction workload. Build a fixture set with a static page, a JavaScript application and any PDFs your pipeline depends on. Pin the Webclaw version, record whether each case uses HTTP, a local browser or the hosted API, and compare the extracted content with the source. For MCP, also test initialization, tool listing and a real scrape with your intended client.

