MCP server

Knowpoint has an MCP server, so the AI tools your team already uses can look things up in your documentation: "What's the VPN setup at Contoso?", "Which servers does Harborview have?", "Write a runbook for the new firewall."

It works with any client that supports the Model Context Protocol (MCP), including Claude Code, Claude Desktop, Cursor, VS Code with GitHub Copilot, Windsurf and Zed.

#What it can do

The AI gets the same tools Knowa uses inside Knowpoint:

• Search and read: organizations, assets, flexible assets, documents and the knowledge base, contacts, locations, domains, tags, and credentials (names and usernames only).
• Make changes, only if you allow it: create and update documents, assets, flexible assets, contacts, locations and organizations, link related items and add tags.

Password values are never shared, whatever the API key allows. The AI runs outside Knowpoint, so secrets stay behind the Reveal button. If someone asks for a password, the AI points them to it in Knowpoint.

#What it can see

Everything goes through an API key, which acts as a person:

• It can never see or do more than that person can: their role, the organizations they can access, and items restricted to other people.
• The key's own permissions narrow it further: which resources (read, or read and write), which organizations, and which asset and flexible asset types. Tools the key doesn't allow aren't offered to the AI at all.
• Changes are recorded in the audit log, like any other.

For most people, a read-only key is the right choice.

#Set it up

1. Go to Admin → AI & API and select + New API key. Choose who it acts as and what it can do, for example Read for everything except passwords. Copy the key.
2. On the same page, the MCP server card shows your server address, https://<your workspace>.knowpoint.io/mcp, and ready-to-paste setup for each client, with your new key filled in.
3. Add it to your AI tool:

Claude Code

claude mcp add --transport http knowpoint https://<your workspace>.knowpoint.io/mcp --header "Authorization: Bearer <your API key>"

Cursor (~/.cursor/mcp.json) and other clients that take a URL and headers:

{
  "mcpServers": {
    "knowpoint": {
      "url": "https://<your workspace>.knowpoint.io/mcp",
      "headers": { "Authorization": "Bearer <your API key>" }
    }
  }
}

VS Code with GitHub Copilot (.vscode/mcp.json): the same, under "servers", with "type": "http".

Claude Desktop: in Settings → Developer → Edit Config, add the server with mcp-remote (it needs Node.js). The exact snippet is on the MCP server card. Restart Claude afterwards.

4. Ask a question that needs your documentation. The first time, your client may ask you to approve the Knowpoint tools.

#Good to know

• Each person should use their own key, so the audit log shows who did what and access stays personal.
• Revoke a key in Admin → AI & API to cut off that connection immediately.
• IP allowlist: if your workspace has one and it covers the REST API, it covers the MCP server too.
• Limits: up to 300 requests a minute per key.
• Technical details: Streamable HTTP transport at /mcp, JSON responses (no event stream, no sessions), MCP protocol versions 2024-11-05 to 2025-11-25.

#Troubleshooting

• 401 or "Invalid API key": check the key was copied in full and hasn't been revoked, and that the header is Authorization: Bearer <key>.
• "This API key doesn't allow that": an admin can change the key's permissions in Admin → AI & API.
• The AI says it can't see a client: the key, or the person it acts as, doesn't have access to that organization.