Overview
ondoki includes a Model Context Protocol (MCP) server that lets external AI agents search and read your documentation. Connect Claude Desktop, Cursor, GitHub Copilot, or any MCP-compatible tool to your ondoki knowledge base.How It Works
The MCP server is mounted at/mcp on your ondoki backend and uses stateless HTTP transport via FastMCP. AI agents authenticate with project-scoped API keys and can list, search, and read your content.
Available Tools
The MCP server exposes these tools to connected AI agents:| Tool | Description |
|---|---|
list_projects | List accessible projects |
search_pages | Full-text + semantic search on documents |
search_workflows | Full-text search on workflows |
list_workflows | List workflows in a project |
read_document | Get full document content |
read_workflow | Get workflow details with all steps |
read_folder | List folder contents |
create_context_link | Create URL/app → resource associations |
Setup
1. Generate an API Key
In the ondoki UI, go to Project Settings → Integrations and create a new MCP API key:- Click Create API Key
- Enter a name (e.g., “Claude Desktop”)
- Copy the generated key (it’s only shown once)
POST /api/v1/mcp/projects/{project_id}/mcp-keys
2. Configure Your AI Agent
- Claude Desktop
- Claude Code
- Cursor
- Environment Variable
Add to your Claude Desktop MCP config (
claude_desktop_config.json):3. Verify
Ask your AI agent to search your ondoki knowledge base:“Search ondoki for deployment workflows”The agent should use the
search_workflows or search_pages tool to query your content.
Authentication
MCP requests are authenticated via API key:- Header:
Authorization: Bearer ondoki_<key> - Environment:
ONDOKI_API_KEY=ondoki_<key>
- Project-scoped (only access content in the associated project)
- Hashed with SHA-256 in the database
- Trackable (last used timestamp)
- Revocable (can be deactivated)
Managing API Keys
| Action | API |
|---|---|
| List keys | GET /api/v1/mcp/projects/{project_id}/mcp-keys |
| Create key | POST /api/v1/mcp/projects/{project_id}/mcp-keys |
| Revoke key | DELETE /api/v1/mcp/projects/{project_id}/mcp-keys/{key_id} |
Audit Logging
All MCP access is tracked in the audit log with action typeMCP_ACCESS. This includes:
- Which API key was used
- What resources were accessed
- Timestamps and IP addresses
Use Cases
- Claude Code — let Claude search your internal process docs while coding
- Cursor — surface relevant workflows when writing related code
- Custom agents — build AI assistants that can reference your team’s knowledge base
- RAG pipelines — use ondoki as a retrieval source for custom AI applications