Lyro MCP Server

Lyro runs its own MCP (Model Context Protocol) server, so you can connect an AI client - like Claude, Cursor, or ChatGPT - directly to your workspaces. Once connected, the client can read and change your Lyro setup the same way you would in the dashboard: create agents, add knowledge, configure channels, review conversations, and more.

Not to be confused with Tools and integrations, which is about connecting external MCP servers into Lyro as agent tools. This page is the other direction - connecting a client to Lyro's own MCP server.

Connect by signing in (recommended)

Add the server by URL alone - no key to paste:

Endpoint: https://mcp.getlyro.ai/mcp
Transport: Streamable HTTP
Auth: OAuth - your client opens the browser, you sign in to Lyro and pick which workspaces the client may reach

Per client:

Claude Code:
```
claude mcp add --transport http lyro https://mcp.getlyro.ai/mcp
```
Then start Claude Code, run /mcp, pick lyro, and choose Authenticate.
claude.ai: Settings → Connectors → Add custom connector, paste the endpoint URL.
Cursor: add the server to .cursor/mcp.json with just the URL - no headers entry:
```
{
  "mcpServers": {
    "lyro": { "url": "https://mcp.getlyro.ai/mcp" }
  }
}
```
Then choose Login in Cursor's MCP settings to sign in.
ChatGPT: enable Settings → Apps & Connectors → Advanced → Developer mode (paid plans), create a connector with the endpoint URL, and pick OAuth as authentication.

On the first connection the browser opens Lyro's consent screen: you see which app is asking, pick the workspaces it may access, and approve. Reconnecting the same app later skips the consent and just works.

If you granted a single workspace, it becomes the active one automatically. If you granted several, workspace-scoped tools first ask the client to pick one with the select_workspace tool - AI clients handle this on their own, but a hand-written client should call it before anything workspace-scoped.

What a connection can and cannot do

The workspaces you pick on the consent screen are the boundary. Inside them, the client acts as you, with the full tool set - including creating, editing, and deleting things. There is no read-only mode for OAuth connections today, so only connect clients you trust with the workspaces you grant.

A connection can never:

reach a workspace you didn't grant, or one you're no longer a member of
create new workspaces or mint API keys
call any Lyro API outside the MCP endpoint - the tokens are bound to it and rejected everywhere else
manage your other connections

The sign-in scopes a client requests (openid, email) only cover identifying you; what the connection can actually do is decided by the workspaces you grant.

Managing connections

Settings → Connected apps lists every app you've connected, with the workspaces it can reach and when it was last used. From there you can change the workspace set or disconnect an app - a disconnected app loses access on its very next request. Connections are personal: each teammate manages their own.

API keys for unattended use

For CI pipelines, scripts, or anything that can't open a browser, keys still work:

Personal key (krns_uk_...) on https://mcp.getlyro.ai/mcp - spans every workspace you're a member of. Generate it in Settings → MCP.
Workspace key (krns_sk_...) on https://mcp.getlyro.ai/mcp/workspace - locked to the workspace that created it, with optional scopes. Generate it in Settings → MCP under Workspace access; scoped or expiring keys live in Settings → API Keys (see API & Developers).

{
  "mcpServers": {
    "lyro": {
      "url": "https://mcp.getlyro.ai/mcp/workspace",
      "headers": { "Authorization": "Bearer krns_sk_..." }
    }
  }
}

Dashboard session tokens are not accepted on the MCP endpoints - that path is retired. Use OAuth or an API key.

What you can do

With a full-access connection or key, an MCP client can manage:

Agents - create and configure agents, roles, and skills (see Agents)
Knowledge - create knowledge bases and articles (see Knowledge)
Channels - set up the chat widget, email, phone, and other channels
Conversations & contacts - read conversations and look up people
Escalation - manage escalation procedures and destinations
Integrations, custom tools, and MCP servers - wire up the tools your agents can call
Evals & datasets - build and run evaluations
Analytics & findings - pull metrics and quality findings
Workspaces - read and manage workspace settings

A workspace key with limited scopes only exposes the matching tools. Give a key the narrowest scopes it needs.