> ## Documentation Index
> Fetch the complete documentation index at: https://apidocs.neetokb.com/llms.txt
> Use this file to discover all available pages before exploring further.

# NeetoKB MCP

> Manage your knowledge base from an AI assistant in plain language.

The NeetoKB **MCP server** lets your AI assistant work with your knowledge base from plain-language requests. Ask it to find an article, draft a new one, publish a change, share an unlisted link, or invite a teammate, and it works with your NeetoKB workspace for you.

[MCP](https://modelcontextprotocol.io) (Model Context Protocol) is an open standard that connects AI assistants to tools such as NeetoKB. You describe the task; the assistant uses the right tool.

## What you can do

<CardGroup cols={2}>
  <Card title="Find articles" icon="magnifying-glass">
    Search and browse published articles and drafts, and read their full content.
  </Card>

  <Card title="Write and publish" icon="pen-to-square">
    Create articles in a category, revise them, and move them between draft and published.
  </Card>

  <Card title="Share privately" icon="link">
    Generate and regenerate unlisted links, with or without an expiry.
  </Card>

  <Card title="Manage the workspace" icon="user-group">
    Browse the category tree, list authors and recommendations, and invite team members.
  </Card>
</CardGroup>

## MCP vs CLI: which should I use?

NeetoKB's [CLI](/cli/introduction) reaches the same resources MCP does - articles, unlisted links, categories, authors, recommendations, team members, and workspace settings. The one exception is small: only MCP can read a published article from a knowledge base you are not signed in to. Otherwise neither side can do more than the other, so choose on how the work reaches NeetoKB.

### Reach for MCP when

* **The words live in your chat, not in your head.** A support thread, a customer email, or a pasted release note becomes the article body with no retyping and no hand-written markup. The CLI wants the finished HTML from you, and it cannot see the conversation it should come from.
* **You have not decided the steps yet.** "Our refund window moved to 30 days - fix whatever still says 14" means searching, reading each hit, and judging which ones are genuinely wrong. A command can only carry out a decision you have already made.
* **One request should cover several steps.** Find the article, read what it says today, rewrite only the paragraph that is stale, publish it, and hand back an unlisted link for a colleague to check. A script can pass an article id from one command to the next, but it cannot read the body and work out what to change in it.
* **The person doing it does not use a terminal.** A support lead or a writer can keep the knowledge base current from a chat window. NeetoKB hosts the server, so there is nothing to install or keep updated.

### Reach for the CLI instead when

* **No AI assistant should be in the loop.** A cron entry or a docs pipeline runs the CLI against a workspace it is already signed in to - no assistant open, no model account, no tokens spent per run. Every MCP call needs something with model access running.
* **The output feeds another program.** The CLI prints a bare article id, or raw JSON with a `pagination` block, straight into `jq`, a spreadsheet, or your own script. Here you get prose you would have to copy out by hand.
* **You are sweeping the whole knowledge base.** Here every page is a separate tool call, and a thousand article titles crowd out the assistant's context long before the audit is finished. The CLI returns `total_pages` and `total_records` next to the records, so a shell loop walks every page unattended and writes each one to a file or into `jq` - the size of the knowledge base stops mattering.
* **The run has to be repeatable and reviewable.** A command is the artifact: it records exactly what ran and repeats identically in a runbook or a pull request. Ask twice here and the assistant may take a different route.

<Tip>
  You can have both. Run [`neetokb setup claude`](/cli/ai-assistants) and the
  same assistant drives the CLI for you, so a plain-language request here can
  still end in an exact command you can read, repeat, and paste into a
  script.
</Tip>

## What you need

1. A supported AI assistant - **Claude, ChatGPT, Claude Code, Codex, Cursor, Gemini CLI, VS Code (GitHub Copilot), or Windsurf**.
2. A way to authenticate. Claude and ChatGPT sign you in with OAuth, so there is nothing to copy. The others use a NeetoKB **API key** - [learn how to generate one](/mcp/authentication).

<Note>
  The two credentials differ in reach, not just in setup. An OAuth grant runs as **you**, so the assistant sees only what your permissions allow. An API key carries no identity and reaches the **whole workspace**, drafts and private categories included. [Authentication](/mcp/authentication) explains which to pick.
</Note>

[Connect your assistant](/mcp/connect) to get started.