MCP-Native by Default

Every other tracker bolts on an API and calls it "integrations." We built xpntl MCP-first so your coding agents get the same 70+ tools a human uses, with no middleware and no glue code.

The Integration Tax

Typical tracker integrations look like this: a REST API, an OAuth dance, a webhook endpoint you have to host, and a brittle transform layer that maps their data model to yours. Every agent harness reinvents this. Every update breaks something.

MCP (Model Context Protocol) changes the equation. It's a standard protocol that AI agents already speak. Instead of building a custom integration per tracker, the agent connects to a single MCP endpoint and gets structured tools.

70+ Tools, Zero Glue

The xpntl MCP server exposes everything an agent needs:

• issue_create, issue_update, issue_list, issue_get
• comment_create, comment_list
• label_list, label_add, label_remove
• project_list, workflow_state_list
• issue_assign, issue_unassign
• inbox — unread @mentions, assignments, and replies in one call
• doc_list, doc_get, doc_create, doc_update — read and maintain the workspace wiki
• Plus cycles, teams, and loops. Every tool is typed, documented, and streaming.

No REST wrappers. No webhook servers. No token rotation headaches. The agent connects at mcp.xpntl.dev/mcp with a harness key and starts working.

Why This Matters

When your agent can create an issue, assign itself, post progress comments, and move the issue to Done, all through a native protocol, the human overhead drops to near zero. You review the PR, not the tracker bureaucracy.

That's what MCP-native means. Not a published API, not a webhook feed, but your agents working as first-class team members from day one.