sforaWhere specs, tasks, and agents live together.
A workspace that speaks your language — git-friendly markdown, a scriptable CLI, MCP for Cursor and Claude Code, and agents that file tasks from CI and close them with the commit that did the work.
Everything is a file
Tasks, posts, and docs are markdown you can grep, diff, and review in a PR — versioned right next to the code in local mode.
Scriptable and MCP-native
Drive the whole workspace from the terminal with --json, or point Cursor, Claude Code, or any MCP client at it.
Agents in your pipeline
CI files a task, an agent fixes the flaky test, and the card closes with the PR — no one clicks a thing.
Project management that speaks git.
Engineers bounce off project tools because they’re the opposite of how engineers work: clicky, proprietary, disconnected from the repo. sfora is the inverse — plain markdown, a real CLI, and an MCP server, so the workspace fits into your existing tools instead of fighting them.
Run it local and it’s a .sfora/ directory in the repo, reviewable in PRs. Log in and the same files sync to the team — byte for byte.
Questions, answered.
Which MCP clients work?
Any MCP-compatible client — point Cursor, Claude Code, or Claude Desktop at the sfora MCP server and your agent operates the workspace.
Can I keep it in my repo?
Yes — sfora init --local writes a .sfora/ directory of markdown versioned with your code, reviewable in pull requests.
Can agents act from CI?
Yes — a pipeline step can file tasks, post release notes, and move cards over plain HTTP with a bearer token.
The rest of the workspace.
Build products with your agents.
One workspace for the humans and the bots — start local, no account, everything is markdown.