Connect any tool to your project.
An HTTP API that documents itself, an MCP server for any agent client, OAuth 2.1 sign-in and live event streams. Whatever connects acts as a real account, within its roles.
An HTTP API that documents itself, an MCP server for any agent client, OAuth 2.1 sign-in and live event streams. Whatever connects acts as a real account, within its roles.
The REST API does everything the interface does, with JSON in and out. Address a project by its key and a work item by its ref, like PRJ-42.
The server builds its OpenAPI 3.1 document from its own running routes, with the role each call needs and every error code a client may meet.
Claude Code, Codex, Gemini, Cursor, VS Code or any other MCP client connects over Streamable HTTP and gets more than 90 tools, each checked against your role.
Clients connect through OAuth 2.1: they register themselves, you approve in your browser and may confine the client to chosen projects. Disconnect any app later.
An API key acts as you, never more. Confine it to chosen projects and it cannot reach past them, even in an admin's hands. Revoking it closes its open connections.
A Server-Sent Events stream hands your integration project events as they land, filtered to the events you choose. Each event is checked against the reader's role first.
Connecting is the first job to hand your agent. After that, it writes the glue code against your real project, with Manager's API skill as its reference.
Say to your agentManager: Add this MCP server to your config and reload: https://your-manager/api/v1/mcp
The agent edits its own client config. On the first call your browser opens, you approve the connection, and the client keeps its own token.
Say to your agentManager: Write the code for our crash reporter to file a Bug in PRJ over the HTTP API, using the fields PRJ really has.
It loads Manager's API skill, reads PRJ's types and custom fields and writes the request against them. The API key is yours to create, confined to PRJ.
Say to your agentManager: Write a small service that follows the task events in PRJ and posts each new Bug to our team chat.
It builds a client for the event stream, filtered to task events, that reads its key from an environment variable rather than from the code.
Say to your agentManager: Import the open issues from our old tracker's export into PRJ, keeping their hierarchy.
It loads Manager's import skill, maps the export onto PRJ's types and creates the work items with their parents.
Every surface other software reaches Manager through.
Guides that put this feature to work in a real project.
If your agent speaks MCP, it can work with Manager. Connect it to give it access to your projects in Manager.
Collect bugs and feature requests, grade them by severity, and let your team or your agents fix them, for any project that can break.
Let an agent respond first: open and grade the incident, gather context and notify your incident commander, then draft the postmortem.
A script, a client or an agent: whatever connects acts as a real account and may do what its roles allow, no more. A key confined to a project stays there, and revoking it ends its open connections at once.
Get early access