MCP servers

Model Context Protocol (MCP) is how PP Agent reaches business systems: a CRM, an issue tracker, a database, a document store, or an internal API.

Add a server

Open Settings › MCP servers. You can:

• add a server by hand with a command or URL and its environment;
• sign in with OAuth when the server supports it;
• paste a configuration you already use in Claude Desktop or Cursor.

Review the command and arguments before saving. An MCP server runs with the permissions of the user account that started PP Agent.

Choose which agents see it

MCP servers are registered per agent. Giving a server to one agent does not give it to every agent. This keeps a narrow agent narrow, and keeps a large tool catalog out of unrelated chats.

Check status

/mcp shows the status of the servers available to the current agent, including connection errors and the tools each server exposes. Use it first when a tool call fails.

Approvals

MCP tools follow the same approval system as built-in tools. A tool that changes data can be set to ask, be always allowed, or be turned off. Read-only tools can run freely.

A server may expose both, so name the tools explicitly when you tighten a policy.

OAuth

For servers that support OAuth, the app opens the provider’s sign-in page and stores the resulting token in the vault. When a token expires, the app asks you to sign in again rather than silently failing.

Troubleshooting

Symptom	Check
Server shows as disconnected	/mcp for the error, then the command, URL, and environment
Tools are missing	Whether the agent has the server enabled, then restart the server
Sign-in loops	Whether the provider allows the redirect the server expects
A call needs approval every time	The tool’s policy in Settings › Tools, or the agent’s settings

For the internal request flow, see Architecture and the repository’s providers and MCP document.