Back to Learn

Help

MCP for documentation: serve IDE agents without scraping

Model Context Protocol (MCP) lets IDE agents query your docs as structured tools instead of scraping HTML. ADN serves MCP endpoints for the developer_agent class.

When a developer asks Cursor or Claude Code about your API, the agent does what any agent does: it tries to read your docs. If your documentation is a JavaScript-heavy docs site, the agent scrapes HTML and hopes for the best. Model Context Protocol (MCP) is a better contract.

What MCP gives IDE agents

Instead of fetching pages, agents call tools:

search_docs(query) — full-text search across your documentation

get_page(slug) — retrieve a specific page as clean Markdown

list_endpoints() — API reference index

get_schema(type) — structured type definitions

Tools return exactly what the agent needs. No navigation chrome. No sidebar HTML. No token waste.

How ADN serves MCP

ADN classifies requests from developer_agent traffic (Cursor, Claude Code, Copilot, Windsurf) into the developer_agent intent class. For connected domains, the gateway exposes MCP tools backed by your ingest pipeline:

1. Ingest — ADN crawls your docs (or accepts a push) and normalizes content

2. Index — Pages become searchable Markdown with metadata

3. Serve — IDE agents call MCP tools at your gateway hostname

Humans still browse your docs site normally. Agents get the tool interface.

MCP + llms.txt + Markdown

MCP is the richest shape, but not every agent supports it yet. ADN layers discovery surfaces:

llms.txt — curated link index for agents that browse

Markdown pages — clean content for agents that fetch URLs

MCP tools — structured queries for IDE agents that support the protocol

The classifier picks the best available shape per visitor.

Why docs companies care

For developer tools companies, incorrect answers in the IDE are worse than no answer. A hallucinated API parameter costs your support team a ticket.

Serving MCP means:

Agents quote your actual docs, not a stale blog post

Token cost drops vs scraping full HTML pages

You control rate limits and terms per agent class

Get started

1. Read the docs segment guide for the full ROI argument

2. Run the efficiency scan at https://www.agent-delivery.network/#scan on your docs URL

3. Connect your domain in the ADN dashboard and enable the ingest pipeline

IDE agents are already looking for your docs. Give them a front door that works.

Frequently asked

What is MCP?
Model Context Protocol is an open standard for AI agents to call structured tools — search docs, fetch API schemas, run queries — instead of scraping web pages.
Do I need to build my own MCP server?
ADN can expose MCP tools generated from your ingested documentation. You connect a domain, run the ingest pipeline, and IDE agents get tools at your gateway hostname.

See how this applies to your stack: Docs & dev tools guide.

Continue reading

Run the free efficiency scan on your URL to see what machines receive today.