MCP Gateway

The Lovelace platform exposed through a single MCP gateway

Connect any MCP-compatible client to Lovelace through the public MCP gateway at https://mcp.uselovelace.com/mcp. One endpoint gives your AI client or application access to Lovelace workspaces, agents, and knowledge.

What you can build

Connect your AI client — Add Lovelace to Claude Desktop, Claude Code, Cursor, or Windsurf with a single config entry. Your AI client can then list workspaces, spawn agents, and query knowledge without any code. See Client Setup.

Orchestrate agents from code — Use the Python or TypeScript MCP SDK to spawn Lovelace agents, poll for results, and build fully automated workflows. See Agent Workflows.

Use Lovelace as AI memory — Store documents in a workspace, search them semantically, and pass retrieved context to an agent. See Tools Reference.

How it works

The gateway uses Streamable HTTP transport and MCP authorization discovery:

Endpoint: https://mcp.uselovelace.com/mcp
Messages: POST /mcp
Auth discovery: 401 challenge with resource_metadata
Protected Resource Metadata: https://mcp.uselovelace.com/.well-known/oauth-protected-resource
Session header: mcp-session-id
Session termination: DELETE /mcp

Standards-compliant clients discover Lovelace Accounts through the gateway's Protected Resource Metadata, complete OAuth authorization, and then send bearer tokens on MCP requests. Clients normally handle both the auth flow and the session header automatically.

Quick start

Add this to your MCP client configuration:

json

{
  "mcpServers": {
    "lovelace": {
      "type": "streamable-http",
      "url": "https://mcp.uselovelace.com/mcp"
    }
  }
}

Then verify the connection by asking your client:

text

List my Lovelace workspaces

See the Quick Start guide for the full end-to-end flow, including programmatic Python and TypeScript examples.

Capabilities

Capability	MCP surface	Description
Workspaces	`lovelace_list_workspaces`, `lovelace_get_workspace`, `lovelace://workspaces`, `lovelace://workspaces/{id}`	Browse and inspect accessible workspaces
Agents	`lovelace_list_agents`, `lovelace_spawn_agent`, `lovelace_get_agent_status`, `lovelace_get_agent_result`	Launch agents, monitor progress, and fetch output
Knowledge	`lovelace_search_knowledge`, `lovelace_store_knowledge`, `lovelace://knowledge/{id}`	Search and store workspace knowledge

Scopes

Each tool requires a specific OAuth scope. Request the scopes you need when your client completes the authorization flow.

Scope	Tools unlocked
`workspace:read`	`lovelace_list_workspaces`, `lovelace_get_workspace`
`agents:read`	`lovelace_list_agents`, `lovelace_get_agent_status`, `lovelace_get_agent_result`
`agents:write`	`lovelace_spawn_agent`
`knowledge:read`	`lovelace_search_knowledge`
`knowledge:write`	`lovelace_store_knowledge`

Documentation

Quick Start — Configure a client, verify access, spawn an agent, and fetch results
Agent Workflows — End-to-end Python and TypeScript examples for spawning and orchestrating agents
Authentication — Protected-resource discovery, OAuth, and scope guidance
Tools Reference — Current tool inputs, outputs, and required scopes
Compatibility Report — Per-client support matrix generated from conformance results
Capability Lifecycle — Deprecation metadata, replacement handling, and removed-tool errors
Resources Reference — Current lovelace:// resource URIs and payload shapes
Client Setup — Claude Desktop, Claude Code, Cursor, Windsurf, and generic clients
Local Server — Run the local MCP server through the daemon/Ada CLI

Lovelace Developer Portal