Troth documentation
Troth is a structural map of a codebase. Specs are areas of the code and they persist as long as those areas exist. Tasks are units of work inside a spec. The tree of specs and sub-specs is meant to look like the architecture of the system it tracks, and the two evolve together.
It is not a project tracker, a wiki, a kanban board, or a sprint planner.
There are two ways in.
The CLI. troth runs in your terminal next to the agent doing the work. It initializes a project, decomposes specs into tasks, hands tasks to Claude Code, and runs the verification gates that decide whether a task passed. Start at Getting started.
The MCP server. A hosted Model Context Protocol server at https://mcp.troth.dev/mcp gives Claude and other MCP clients direct access to the same project. Use it when you are planning in a chat window rather than executing in a terminal. Start at Connect Troth to Claude.
Both read and write the same cloud project, so a spec written in a conversation is the spec the terminal sees.
Where to go next
Section titled “Where to go next”| Page | What it covers |
|---|---|
| Getting started | Install, log in, initialize a project, run your first task |
| Connect Troth to Claude | Step-by-step MCP setup for Claude.ai, Claude Code, and Cursor |
| Concepts | Specs, tasks, requests, findings, foundations |
| CLI reference | Every command, generated from the source |
| MCP tool reference | Every MCP tool, generated from the source |
The two reference pages are generated from the Troth source on every build, so they describe the version that is actually shipping.