Web Fetch & HTTP Requests

Two URL tools

The web toolset has two tools that act on a URL you already have. (To discover a URL first, use web__web_search, covered separately.)

web__web_fetch reads a human-facing web page or document and returns clean markdown of its main content, with navigation, sidebars, and scripts stripped out.
web__http_request performs an HTTP request and returns the raw response status, headers, and body as JSON. Use it for JSON/API endpoints, webhooks, or when you need to inspect raw status and headers.

Pick web_fetch to read a page and http_request to call an endpoint.

web_fetch

Fetches a URL and returns markdown of the page's main content. The output is a small header (# <title> when the page has one, plus a Source: <final-url> line) followed by the extracted markdown. When extraction yields very little text, primer appends a note that the page may require JavaScript rendering and suggests configuring a JS-capable web-fetch provider.

Parameter	Required	Type	Notes
`url`	Yes	string (http/https)	Absolute URL of the page to read.
`max_chars`	No	integer > 0	Cap the returned markdown to this many characters.
`max_lines`	No	integer > 0	Cap the returned markdown to this many lines.

Machine metadata (HTTP status, content type, final URL after redirects, character and line counts, and whether the output was truncated by a limit or judged thin) is returned alongside the markdown as extended fields.

Web fetch routes through a configured web-fetch provider, so behaviour and JavaScript support depend on which provider is active.

http_request

Performs a single HTTP request and returns a JSON object with the response.

Parameter	Required	Type	Notes
`url`	Yes	string (http/https)	Absolute URL to request.
`method`	No	enum	One of GET, POST, PUT, PATCH, DELETE, HEAD, OPTIONS. Defaults to GET.
`headers`	No	object	Optional request headers; keys and values are strings.
`body`	No	string	Optional request body. Serialise structured payloads (JSON, form encoding) yourself before the call.
`timeout_seconds`	No	number	Per-request timeout, greater than 0 and at most 300. Defaults to 30.

The response is JSON with four fields: status (the HTTP status code), headers (the response headers), body (the response body as text), and truncated (a boolean).

Output truncation

The response body is capped at a configurable byte limit. The default cap is 1 MB (1,000,000 bytes). When the body exceeds the cap it is truncated to the cap and truncated is set to true, so the agent can tell that more content was available. The cap is configured per deployment.

Configure the web search providers that back web_search, the discovery counterpart to these two tools.

Exposing web tools (and others) to external clients when primer runs as an MCP server.