MCP server
Connect Claude, Cursor or any MCP client, and the tools it gets.
The MCP server lets an AI agent work in an organization with the role of the member who connected it. It is at:
https://example.com/mcp
Connect an agent
An agent that can open a browser signs in through OAuth: the member picks the organization and allows read, or read and write. An agent that cannot open a browser sends an API key that holds the mcp scope.
claude mcp add --transport http acme https://example.com/mcp
{ "mcpServers": { "acme": { "url": "https://example.com/mcp" } } }
{ "servers": { "acme": { "type": "http", "url": "https://example.com/mcp" } } }
Tools
| Tool | Does | Needs |
|---|---|---|
get_organization | The plan, the billing state and the credits. | read |
get_balance | The credits left and the next ones to expire. | read |
get_usage | Calls and credits over a range. | read |
list_keys | The API keys, never their secrets. | read |
list_items | The items, 50 at a time. | read |
get_item | One item. | read |
create_item | Creates an item, for 1 credit. | write |
update_item | Renames, describes, archives or restores an item, for 1 credit. | write |
search_docs | Searches these docs. | read |
get_doc_page | One page of these docs, in Markdown. | read |
A write takes an optional request_id, so that a retried call is charged and written once. Its credits are counted under the agent in the usage page.
Control the agents
AI agents in the dashboard lists the connected agents and the calls they made. An owner or an admin can disconnect one agent, or stop every agent with one switch.