Documentation MCP

This documentation is also served as a public MCP server. Connect the endpoint to your AI client and any question about Hilbana gets answered from the actual pages on this site, rather than from whatever the model remembers from training.

It is read-only and needs no key: everything it returns is already published here.

Not the same as the app’s MCP

Two different servers, two different jobs:

	Documentation MCP	App MCP
Endpoint	https://hilbana.com/mcp	https://app.hilbana.com/mcp
Authentication	None	hil_… key or OAuth
What it does	Read the documentation	Operate your issues, projects and documents
What it’s for	Learning and answering questions	Working

If what you want is an agent that creates issues, pulls from the queue or changes states, you need the other one: see API & MCP.

Connecting it

In Claude Code, one line:

claude mcp add --transport http hilbana-docs https://hilbana.com/mcp

In claude.ai (web and desktop), add it as a custom connector with that same URL: Settings → Connectors → Add custom connector.

If your client is configured with a JSON file (.mcp.json, claude_desktop_config.json, your editor’s mcp.json…), here is the equivalent entry. Nothing to replace, copy it as is:

{
  "mcpServers": {
    "hilbana-docs": {
      "type": "http",
      "url": "https://hilbana.com/mcp"
    }
  }
}

The tools

Tool	What it returns
search_docs	The most relevant pages for a query, with excerpts from the text
get_doc	The full Markdown of one or more pages, by slug
list_docs	The index: slug, title, description and section of every page
get_all_docs	All of the documentation for one language, in a single block
get_overview	Product summary and site map: what it is, plans and pricing, links

They all take lang, which defaults to es and accepts en: the documentation exists in full in both languages.

The surprising one is get_all_docs. The complete documentation for one language is around 30 pages, which fits comfortably in a current model’s context, so your client can ask for everything at once and answer without chaining searches. If you’d rather go piece by piece, search_docs locates and get_doc reads.

What to ask it

Once connected, it covers what you’d do with the documentation open next to you, without opening it:

• “How do I connect an agent to Hilbana from Claude Code?”
• “Explain the difference between cycles and milestones in Hilbana.”
• “What does claim_issue actually do, and why call it before starting work?”
• “According to Hilbana’s documentation, how do I sign an inbound webhook?”
• “Summarise what the free plan includes.”

How it stays current

The server’s content is generated in the same deployment as this site, from these very files. There’s no separate index to sync and no copy that can go stale: if you can read it here, the MCP returns it.