NeetoKB MCP server

What is NeetoKB MCP?

NeetoKB MCP lets AI assistants like Claude and ChatGPT work with your NeetoKB knowledge base. Once it's connected, you can ask your AI assistant what your help center already covers, draft an article from a support conversation, clean up outdated content, and publish, in the same conversation where you're already working.

MCP stands for Model Context Protocol, an open standard for connecting AI assistants to the apps you use. You don't need to know anything about it. You connect once, sign in to NeetoKB, and start asking.

Your AI assistant talks to the NeetoKB MCP server. NeetoKB hosts it, so there's nothing to install or keep updated, and every workspace uses the same address:

https://connect.neetokb.com/mcp/messages

When you connect, you sign in with your NeetoKB account and pick your workspace. The assistant then works as you: it sees the articles and categories you see in NeetoKB, and does only what your role allows. It can search and read articles, create, update, schedule, and publish them, restore old revisions, add images and files, trash and restore articles, and share them with private links. It can also manage your categories, tags, recommendations, redirections, navigation links, settings, themes, and webhooks, read your search and visit reports, and manage team members. You never name a tool; you ask in plain language. The full list is in the NeetoKB MCP tools reference.

NeetoKB also has a command-line tool, the NeetoKB CLI. It works with the same articles, but runs in your terminal, in scripts, in CI, and in scheduled jobs. It's open source, so you can read exactly what it does, change it, and build on it.

Connect any MCP client

Every client uses the same server URL, https://connect.neetokb.com/mcp/messages, over HTTP (Streamable HTTP). Sign in with OAuth unless the client only takes an API key.

Step-by-step guides for each AI assistant:

• Claude
• ChatGPT
• Claude Code
• Cursor
• Codex
• VS Code
• Antigravity CLI
• Devin
• Antigravity

Which sign-in each client supports

Sign in and choose permissions

Every NeetoKB workspace has its own subdomain. When you sign in, enter your workspace subdomain, not your email address. For acme.neetokb.com, enter acme.

What each permission allows

If you leave a permission unticked, a tool that needs it is refused with a message that names the permission. A read-only connection can search and read articles but cannot create, publish, or delete anything.

For developers: the OAuth metadata is at https://connect.neetokb.com/.well-known/oauth-authorization-server. The server supports the authorization_code and refresh_token grants with PKCE (S256) required, and the scopes read, write, delete, and offline_access. Clients register dynamically, per RFC 7591, at https://connect.neetokb.com/mcp/oauth/register, and loopback redirect URIs are accepted. Refresh tokens are issued, so you stay signed in.

Connect with an API key

An API key is not tied to a person. Every tool call reaches the whole workspace, and role checks do not run. Use a key for automation that must see every article, including drafts and private categories. Do not use one on a machine where a person should see only their own articles.

It is the same key the REST API uses. Learn how to generate your API key. Send it in this header:

Authorization: Bearer YOUR_API_KEY

Each AI assistant's guide shows where its config file lives and how it sends the key.

Things to know about API keys

• One key is one workspace. To reach two workspaces, add the server twice, with a different key and a different server name in each entry.
• Keep the key private. It gives access to every article in the workspace. Keep it out of shared config files and commits, and revoke it in your workspace settings if it leaks.

How each interface signs in

NeetoKB MCP tools

Once you connect, your AI assistant lists every tool and what it does. You never need to name one. This is a summary of what the tools cover.

• Articles. Search, read, create, edit, clone, archive and unarchive articles, discard a draft, and update many articles at once. Drafts and published articles are both included.
• Trash. Move articles to the trash, see how many days are left before each is deleted for good, restore them, or delete them permanently.
• Schedules. See, add and remove scheduled publish and unpublish times for an article.
• Article SEO. Read and change the SEO settings of one article.
• Revisions. See an article's earlier versions, read one, and restore it into the draft.
• Attachments. Upload a file or an image and use it in an article.
• Unlisted links. See an article's secret share link, or issue a new one that replaces the old one.
• Categories. See, create, rename, move, reorder and delete categories, and set the order of the articles in a category.
• Tags. See, create, edit, delete and merge tags.
• Authors. See the people credited on articles, with article counts.
• Recommendations. See, create, change and delete the article recommendations shown on matching pages.
• Redirections. See, create, change and delete page and domain redirections.
• Navigation links. See and replace the header and footer links of the public knowledge base.
• Reports. Read popular searches, searches that found nothing, invalid URLs and daily visits. Clear that data for a period, or all of it.
• Team members. See, invite, update and deactivate team members.
• Settings. Read and change the knowledge base settings, the knowledge base SEO settings, your own preferences and the NeetoChat integration.
• Themes. See, clone, rename, configure, publish and delete themes, and add, change and remove theme fonts.
• Webhooks. See, create, change and delete webhooks, and see the events a webhook can listen to.

What to know before you connect

• Permanent deletes. Deleting a trashed article, deleting a category, tag, redirection, recommendation, theme or webhook, and clearing report data cannot be undone. Trashed articles can be restored until they are deleted for good, and deactivated team members keep their articles.
• Unlisted links. Only published articles can have an unlisted link. A new link replaces the old one, so any URL you shared before stops working at once. Treat the URL as sensitive.
• Drafts. The assistant cannot discard the draft of an article that is not published, because that draft is its only copy. Trash the article instead.
• Themes. Built-in themes cannot be changed, so clone one first. The published theme cannot be deleted.
• Who sees what. With OAuth, the assistant sees what the person who approved the connection can see, and a record they cannot open is refused. An API key reaches the whole workspace.
• Roles. With OAuth, the assistant also follows your NeetoKB role. Editing articles, publishing, trashing, restoring, and managing categories, tags, recommendations, redirections, webhooks, settings, themes and team members each need their own role permission, and changing a role needs the manage-roles permission. These are separate from Create and update and Delete. An API key skips role checks.
• Your preferences. Reading and changing your own preferences needs a signed-in user, so an API key connection is refused.
• Public articles. The assistant can read a published article from any workspace that does not require sign-in. It never sees drafts, private categories or the author's email address.

Troubleshooting

Your AI assistant shows no NeetoKB tools

Restart the client first. Every client reads its MCP config at startup. On Claude or ChatGPT, confirm that you added the connector and finished the sign-in. In ChatGPT, the plugin is created only after you tick I understand and want to continue.

On clients that read a config file, check these:

1. The config key. Claude Code, Cursor, Devin, Antigravity, and Antigravity CLI use mcpServers. VS Code uses servers. Codex uses a [mcp_servers.<name>] section. A server under the wrong key is ignored silently.
2. The URL key. Antigravity and Antigravity CLI expect serverUrl. Claude Code, Codex, Cursor, Devin, and VS Code expect url.
3. The endpoint. It is https://connect.neetokb.com/mcp/messages, with no trailing slash.
4. The syntax. In JSON, a trailing comma or an unclosed brace makes the whole file unreadable, and other servers stop working too.

The connection is refused or unauthorized

With an API key, the Authorization header must read Bearer YOUR_API_KEY, with the word Bearer and one space before the key. Check that the key is not revoked and belongs to the workspace you expect. Generate a new key from your workspace settings if you are unsure.

With OAuth there is no key to check. Remove the connector and add it again to start the sign-in over.

The client sends a key when you meant to sign in

A client that finds a credential in its config never starts the sign-in. Remove the headers block, or the bearer_token_env_var line in Codex, and restart the client.

The client asks you to sign in when you meant to use a key

The credential is missing, so the client falls back to OAuth. Add the headers block with "Authorization": "Bearer YOUR_API_KEY", or set bearer_token_env_var = "NEETOKB_API_KEY" in Codex and export that variable. Then restart the client.

The sign-in fails at the workspace step

The first screen needs a workspace subdomain or a workspace URL, not an email address. For acme.neetokb.com, enter acme.

Your AI assistant sees fewer articles than expected

This is how OAuth works. The assistant acts as the person who approved the connection, so lists and searches show only what that person can see. If the assistant must reach everything in the workspace, use an API key.

Your AI assistant reaches the wrong workspace

With OAuth, name the workspace in your prompt. To add another one, sign in again and tick it. An API key belongs to one workspace, so add the server a second time with the other key and a different server name.

VS Code never offers the tools

MCP needs VS Code 1.99 or later with GitHub Copilot in Agent mode. It is not available in Ask or Edit mode. Check the version, switch Copilot to Agent mode, and confirm that .vscode/mcp.json is in the workspace you have open. With an API key, VS Code asks for the key the first time the server is used, so answer that prompt.

Devin or Antigravity connects but the tools are missing

Enable the server after saving the config. In Devin, open the ... menu at the top of a Devin chat in the IDE view and turn the server on, then run devin mcp login <name>. In Antigravity, go to Settings → Customizations.

Claude shows the connector as unauthorized

Open Customize → Connectors, remove NeetoKB, and connect it again from the NeetoKB listing. On a Team plan, Requested means an admin has not turned NeetoKB on yet. An admin turns it on under Organization settings → Connectors.

Check that the server is reachable

The server publishes its OAuth metadata without any credential:

curl -s https://connect.neetokb.com/.well-known/oauth-authorization-server

A JSON document whose issuer is https://connect.neetokb.com means the host is answering, so look at the client's setup next. It does not prove that every tool call works. No answer, or an error page, means the server cannot be reached from your network. Try again in a few minutes, and if it persists, tell us.

Still stuck

Reach us through the help center, or at support@neetokb.com. Tell us which client you are using and what your AI assistant reports when it tries a NeetoKB tool.

Frequently asked questions

Does NeetoKB MCP require an API key?

No. Sign in with your NeetoKB account and approve access. There's no API key to copy or manage.

Any AI assistant that can send an Authorization header, such as Claude Code, Cursor, Codex, VS Code, Devin, and Antigravity CLI, can use a NeetoKB API key instead. Claude and ChatGPT use sign-in only. However, API keys provide access to the entire workspace rather than operating with your individual permissions.

Where is the NeetoKB MCP server hosted?

NeetoKB hosts the MCP server at https://connect.neetokb.com/mcp/messages. You don't need to install, configure, or maintain a server yourself.

Use NeetoKB in other AI assistants

Prefer the terminal? The NeetoKB CLI reaches the same data from your shell, your scripts and your coding agent.