DocsAlot takes scattered help centre articles, knowledge bases and developer docs, folds them into one source of truth, and serves that source to AI agents through hosted MCP, llms.txt and skill.md. It launched on Product Hunt with a one-line pitch that doubles as a description of how documentation work now happens: maintain your help centre by asking Claude, Cursor or Codex. For anyone running a desktop agent against their own product’s docs, the interesting part is not the product, it is the packaging — three different delivery mechanisms for three different jobs.
Key facts
| Fact | Detail |
|---|---|
| Question answered | DocsAlot Bundles MCP, llms.txt and skill.md: Documentation Built for Agents to Read |
| Short answer | DocsAlot collects scattered help centre articles, knowledge bases and developer docs into one source of truth, and serves it to AI agents through hosted MCP, llms.txt and skill.md. |
| Reading time | ~3 min |
| Last updated | 2026-10-05 |
The three surfaces, and why they are separate
- MCP (Model Context Protocol) is live tool access. The agent queries the documentation service at the moment it needs an answer, so it sees the current version rather than a snapshot baked into a prompt.
- llms.txt is a curated index: a file a site publishes to tell models which pages matter and what they contain, so a model does not have to guess from search results.
- skill.md is instructions, not facts. It tells a coding agent how to perform a task — the procedure, the conventions, the gotchas — which is a different job from holding current content.
Treating these as one artifact is the common mistake. An index without live access goes stale; live access without instructions means the agent rediscovers your conventions on every run.
Why this matters for a desktop agent
The failure mode this addresses is quiet. An agent asked to answer a support question, write integration code or draft release notes will use whatever it knows — training data, a cached page, or a document someone pasted in six months ago. Nothing errors. The answer is simply wrong, and it looks confident.
That gets worse on a desktop, where the agent also has your files. An agent that misreads a pricing table in a stale doc and then edits the correct one locally has produced a genuinely confusing state.
Practical habits that follow:
- Point agents at maintained sources, not pasted copies. A prompt stuffed with documentation is a snapshot with an expiry date.
- Keep one canonical location. If the answer to “what does this parameter do” lives in three places, the agent will find the wrong one eventually.
- Separate facts from procedures. Current facts go in the served content; stable procedures belong in agent instructions.
- Version the instructions. When you change a workflow, the skill file has to change with it.
Where MOOGH fits
The same principle applies to our own setup: an agent working on your desktop should read the current state of your files and your apps, not a memory of them. MOOGH runs the agent locally so that the work of reading, checking and updating happens against what is actually on the machine. You can Download it and try it on a documentation folder you maintain.
The takeaway
Documentation is now a product surface for two audiences, and the second one does not skim. If your docs are not machine-readable where it counts — current, indexed, and paired with instructions — your agents will behave like someone working from last quarter’s printout.
Related reading: