Launching today
Almanzor
Ask AI + MCP server for your docs, with cited answers
31 followers
Ask AI + MCP server for your docs, with cited answers
31 followers
Paste your docs URL and get an assistant that answers from your pages, cites its source and says when it doesn't know. Add it with one script tag, share a link, or let coding agents search your docs through a hosted MCP server.
Free Options
Launch Team
Render Sandboxes: a computer for every agentRun untrusted code safely. Idle sandboxes cost nothing
Promoted
Hi Product Hunt! I'm Victor, the solo maker of Almanzor.
Docs search usually fails in two ways: keyword search misses the question, and AI chat makes things up. Almanzor is built to avoid both:
Answers only from your docs and links the page it used.
Says "I don't know" when the answer isn't there, and collects those questions for you. That's your to-do list for the docs.
Comes with a hosted MCP server on the same content, so your users' coding agents (Cursor, Claude Code, VS Code) can search your docs directly.
Try it without signing up: ask the assistant on our own docs at almanzor.cloud/docs, or paste your docs URL on almanzor.cloud to get a demo on your content.
Pricing: 14-day trial, no card. Code LAUNCH50 gives 50% off for 3 months to the first 10 paying customers.
I'll be here all day: tell me what breaks on your docs.
Congrats Victor, from another solo maker! Collecting the "I don't know" questions as a to-do list for the docs is a smart touch. Can Almanzor index docs that live as Markdown files in a repo, or does it need a published site to crawl?
collecting the questions it couldn't answer into a to-do list for the docs is the usefull part, today that list is mostly a guess from support tickets
does the MCP server give Cursor the same cited page links or just the text??? and how often does it recrawl when the docs change???
@thamibenjelloun Thanks, that's exactly the gap we built it for: the list comes from real questions people asked, not guesses from tickets.
Citations in Cursor: yes. The MCP tool returns the matching excerpts, each with its page URL and a number, and tells the agent to answer only from them and cite them as [n] with the link. So Cursor gets the same source pages the widget shows, not bare text. When nothing matches, it says the docs don't cover it instead of letting the agent guess.
Recrawling: today it's on demand, not automatic. In the portal you click "Import again" on the source and the new version replaces the old one. You can also trigger an import from the REST API or the management MCP, so you can hook it into your docs deploy. A scheduled re-import is on the list, btw.