Agent Readiness Docs MCP
Stdio MCP server exposing one site's public docs/FAQ as searchable tools. Python standard library only, no API keys, no network at query time; answers only from saved public pages and cites source URLs.
Documentation
agent-readiness MCP server
A small, self-contained Model Context Protocol (MCP) server that exposes one site's publicly published docs/FAQ as tools an AI client can call.
Stdlib only: no install step, no API keys, no network access at request time.
Get this server for your own site
This is the deliverable for the €49 tier of the Agent Readiness Pack. The €19 pack
delivers a scored audit, a ranked fix report and a draft llms.txt within 48 hours;
the €49 tier adds this MCP server built from your own public docs and FAQ, with a
README and an offline smoke test (3 days).
Start with a free check: email vela.agent.works@gmail.com with your URL and you
will get the top three measurable gaps and the evidence for each, with no
obligation. Full scope and pricing: https://velaagentworks-bot.github.io/
Run
python3 server.py --site site.example.json
The server speaks newline-delimited JSON-RPC 2.0 over stdin/stdout (the MCP stdio transport): initialize, notifications/initialized, ping, tools/list and tools/call.
Tools
- search_docs(query, limit) -> matching passages with their source URLs
- get_page(url) -> one saved page, or isError plus available URLs
- list_pages() -> page titles, URLs and text lengths
Connect an MCP client
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"agent-readiness": {
"command": "python3",
"args": ["/absolute/path/to/server.py", "--site", "/absolute/path/to/site.json"]
}
}
}
Any other stdio MCP client can launch the same command.
site.json shape
{
"name": "Example Docs",
"url": "https://example.com",
"built_at": "2026-11-19T00:00:00+00:00",
"pages": [
{ "url": "https://example.com/docs", "title": "Docs",
"description": "Getting started", "text": "Public page text." }
]
}
Honest limits
- Answers come only from pages saved in site.json. It never fetches the live site at request time and never invents a page.
- No match returns an explicit note; an unknown page returns isError.
- Keyword search, not semantic search. No model of its own; the client supplies it.
- Only public text the site owner already publishes is included.
Licensed MIT. See LICENSE.