ContextFetch is a web search and page-to-Markdown API for AI agents. It returns search results with the page text as Markdown, plus highlights and a pruned context: only the chunks of each page that match your query.

n8n is a fair-code licensed workflow automation platform.

Installation · Operations · Credentials · Use it as an AI Agent tool · Usage examples · Errors · Compatibility · Resources

Installation

Follow the installation guide in the n8n community nodes documentation. In short: Settings > Community Nodes > Install, then enter n8n-nodes-contextfetch.

Only Owner and Admin users can install community nodes. This package is not verified by n8n yet, and unverified community nodes are not available on n8n Cloud, so use self-hosted n8n.

Operations

Each operation is one call to the ContextFetch API. Credits are charged by the API, not by n8n.

Search: search the web. Choose what page content each result carries: Full Text (Markdown), Highlights, Pruned Context. Options: include or exclude domains, published after a date. Outputs one item per result. 1 credit per call.
Get Contents: fetch up to 10 URLs as Markdown, optionally with highlights and a pruned context for a query. Outputs one item per URL; a URL that did not load comes back as an item with url and error. 1 credit per page that loads.
Answer: search, then quote up to 3 of the most relevant passages from the top pages, each followed by [n], with citations. Extractive: no text is generated, and answer is empty when nothing relevant is found. Outputs one item. 1 credit.
Research: run up to 5 searches at once (yours, or derived from the question), rank all pages together and return evidence, one Markdown document of quoted passages with numbered sources, plus results. No LLM runs. Outputs one item. 1 credit per search.
Crawl: follow same-site links from a start URL (up to 3 hops) and get every page as Markdown, optionally ranked and pruned to a query. robots.txt is obeyed. Outputs one item per page. 1 credit per page returned.
Map: list a site's URLs from its links and sitemaps without fetching the pages. Outputs one item per URL. 1 credit.

Credentials

You need a ContextFetch API key.

Get a free key with 2,500 credits (one per network, no signup):
```
curl -X POST https://api.contextfetch.dev/v1/keys/free
```
Or buy credits at https://contextfetch.dev/#buy; the key is shown after payment.
In n8n, create a ContextFetch API credential and paste the key into API Key.
Leave Base URL at https://api.contextfetch.dev unless you use another ContextFetch host.

The credential test calls GET /v1/balance, which costs no credits. The key is sent as Authorization: Bearer <key>.

Use it as an AI Agent tool

The node is usable as a tool (usableAsTool). Add ContextFetch Tool to an AI Agent's tools, pick an operation, and let the model fill the inputs with Let the model define this parameter (a $fromAI() expression). For example, for Search:

Query: {{ $fromAI('query', 'A short web search query', 'string') }}
Max Results: 5
Page Content: Pruned Context

Pruned Context keeps the tool output small: the model gets the chunks of each page that match the query instead of whole pages. When a tool call fails, the agent receives the error message; for HTTP 402 that message includes the payment link, so the agent can pass it to you.

Usage examples

Search, then process each result. Search outputs one item per result with url, title, score, and the content you chose (text, highlights, prunedContext). Connect any node after it; it runs once per result.

Read a list of pages. Put the URLs in Get Contents > URLs, separated by commas or new lines, or map them from a previous node: {{ $json.url }}. Each output item has text (Markdown).

Daily digest without an LLM. Schedule Trigger > ContextFetch Research > Convert to File (evidence field, .md). The evidence field is already a Markdown document with numbered sources.

Site inventory. ContextFetch Map with Include Paths set to /docs/, then Crawl or Get Contents on the URLs you keep.

Errors

HTTP errors become node errors with a message and a next step:

402: out of credits. The message includes the payment link (from the API's action.url or X-Payment-URL), and the description lists the credit packs. After payment the success page shows the key; put it in the credential.
401: the key was not accepted. Check the credential, or get a free key.
400 url_not_allowed: only public http(s) URLs can be fetched.
422: the API rejected a parameter; the message names it.
429, 5xx: try again shortly, or turn on Retry On Fail in the node settings. The API refunds paid requests that fail.

Turn on Settings > On Error > Continue to keep a workflow running and get the error as an item instead.

Compatibility

Tested with n8n 2.43.7 (Docker image docker.n8n.io/n8nio/n8n:2.43.7), installed from the packed npm tarball into ~/.n8n/nodes, running all six operations against the live API and as two AI Agent tools. The 402 path was checked against a local server returning the API's 402 body.