/api/v1/mcp with 66 tools. Most mirror the Public API v1 endpoints; a few are MCP-only. Connect Claude Desktop, Claude Code, Cursor, or any MCP-compatible client and manage your workspace through natural language.
The server uses the same X-API-Key authentication as the REST API. Tools returned by tools/list are filtered based on the scopes assigned to your API key. See API Key Authentication for creating keys and assigning scopes.
Client Configuration
Claude Desktop
Add this to~/Library/Application Support/Claude/claude_desktop_config.json:
Claude Code
Cursor
Add this to~/.cursor/mcp.json:
your-tenant-id with your AGO tenant identifier.
Tool Catalog
All 66 tools are listed below. Most map to a REST API endpoint (except the MCP-only tools noted below). Parameters match the auto-generated endpoint pages in the Endpoints sidebar section.Knowledge - Documents
Scope:knowledge:read / knowledge:write
Knowledge - Sources
Scope:knowledge:read / knowledge:write
Knowledge - Update Requests
Scope:knowledge:read / knowledge:write
Knowledge update requests are detected knowledge-base gaps: questions the agent could not answer well. Use these tools to review and triage them.
Conversations
Scope:conversations:read
Tickets
Scope:tickets:read
Analytics
Scope:analytics:read
Agents
Scope:agents_config:read / agents_config:write
Tools
Scope:agents_config:read / agents_config:write
Prompt Templates
Scope:agents_config:read / agents_config:write
UI Resource Templates
Scope:agents_config:read / agents_config:write
Organizations
Scope:accounts:read
Users
Scope:accounts:read
MCP-Only Tools
Two tools are available through MCP but do not have REST endpoints:search_conversations
Full-text search on message content. Returns threads containing messages that match the query.list_feedback
List all feedback across all conversations with filtering. Eliminates the need to query feedback per-thread.Issue type filters use OR logic: setting multiple to
true returns feedback matching any of the selected types.list_conversations, search_conversations, list_feedback, and list_tickets all accept an agent filter. For conversations, search, and feedback, pass an agent UUID or name (exact, case-insensitive), or a comma-separated list to match several agents. list_tickets accepts a single agent name only (case-insensitive substring match) — it does not accept a UUID or a comma-separated list. Use list_agents to find valid agent names. The filter narrows within your current scope and grants no additional access.Protocol
The MCP server implements JSON-RPC 2.0 over HTTP, following the MCP specification.Session Lifecycle
- Initialize — Send an
initializerequest with your client info and protocol version. The server responds with its capabilities and aMcp-Session-Idheader. - Initialized — Send an
initializednotification to confirm. The server responds with202. - Use tools — Call
tools/listto discover available tools, thentools/callto invoke them. Include theMcp-Session-Idheader on every request. - Terminate — Send a
DELETErequest to end the session.
Headers
Include these headers on every request after initialization:Error Codes
Example: curl Session
Step 1 — Initialize:Mcp-Session-Id header from the response.
Step 2 — Confirm initialization:
content array:
