Built-in tools

These are registered by the MCP server even without optional tool packages. Their presence in list_tools is not proof that their dependencies are installed. They are not exposed as axm.tools entry points by axm-mcp.

verify

Signature: verify(path="."). It calls discovered audit and init_check, then optionally enriches eligible audit failures with ast_impact. It returns text through MCP, with structured audit/governance sections available on the underlying ToolResult.

Absent provider → null section; failed provider → error section. Completed aggregation returns outer success even with quality findings. See Use verify for interpretation and operational effects.

web_fetch

Signature: web_fetch(url, mode="auto").

Mode	Backend behavior
`auto`	Exactly the basic route; no automatic escalation
`basic`	Scrapling Fetcher.get on a worker thread
`dynamic`	Scrapling DynamicFetcher.async_fetch
`stealth`	Scrapling StealthyFetcher.async_fetch

Install the optional backend in the same server environment. The web extra declares scrapling, but the fetcher import requires the backend's fetcher dependencies. The tool's missing-backend diagnostic recommends:

Bash

uv pip install --python /absolute/path/to/server-venv/bin/python "scrapling[fetchers]"

Browser modes additionally require the browser runtime expected by your Scrapling installation. Neither forge nor all includes this extra.

MCP call parameters:

JSON

{"name": "web_fetch", "arguments": {"url": "https://example.com", "mode": "basic"}}

On successful fetching, data contains url, title, text, status_code (possibly null), and the requested mode. Text is cut at 50,000 characters and then receives a truncation marker. Success means a page object was processed, not an HTTP 2xx check: inspect status_code. This tool returns extracted text, not a DOM or Markdown conversion.

Missing backend and invalid mode return failures; fetching exceptions return their error string plus URL/mode. The missing-backend check precedes mode validation, so it can mask an invalid mode. It has no tool-level custom timeout or retry parameter. Dynamic/stealth fetching is not a guarantee of bypassing a site's protections.

Fetching sends a network request and can start a browser. Use it only for URLs you intend the server to contact; this adapter does not implement a network allowlist or SSRF boundary.

list_tools

Call directly with arguments={"kwargs": {}}. The Python body has no required named input, but MCP 1.30 exposes its **kwargs as a required object; omitting it produces a schema validation error. It returns sorted text for discovered tools plus built-in/meta-tool descriptions, including entries not individually exposed by MCP tools/list. It does not execute each tool to check health. The facade meta-tools are described in Facade and result contracts.