Documentation

Understand Tekk

Tekk is your agent for self-improving software — it keeps a spec board, runs autonomous improvement loops on your codebase, and proposes work for you to greenlight. These docs explain the mental model, how to connect your own tools, and how to drive Tekk from inside your coding agent.

Connect

MCP server Connect Tekk to Claude Code, Codex, or Cursor and work your spec board from inside your coding agent — list specs, implement them, run a fleet in parallel, review proposals, read context.

Getting started

What Tekk is The one-page mental model — loops that watch your codebase, propose improvements for you to greenlight, and keep a board of specs your coding agent builds. Core concepts The Tekk vocabulary defined once — workspace, project, spec, goal, proposal, and loop — and how they fit together.

How it works

How the loops work The autonomous side of Tekk — what triggers a loop, how a proposal becomes a spec, and where you stay in control. Loops and their skills The twelve loops Tekk can run on your project, the skills each one carries, and how a run picks the one method it will work with. Integrations Connect your own SaaS tools through Composio so Tekk can see production signals and act on them — Sentry first, more to follow. Pull requests Link GitHub pull requests to your specs so a spec shows what shipped it — and let a merge move the spec to done on its own, gated by its own checklist.

Let your product improve itself.

Connect your repo. Twelve loops start watching your code and bring you proposals — you greenlight them, your coding agent builds them.