A local MCP server that runs a timeboxed ship sprint inside Cursor, Claude Code, or Claude Desktop. Your agent builds; the bunny paces you, gates scope creep, grills your idea, and keeps the ship log.
Habitat-bunny has well-structured tool definitions with clear, domain-specific descriptions (avg 150 - 250 chars). All 8 tools have explicit schemas with typed parameters and enums where appropriate. Naming follows verb_noun convention (start_sprint, lock_idea, check_scope, sprint_status, next_hop, ship, roast, ship_log). However, several tools lack output schema documentation, and error handling guidance is minimal. Parameter descriptions are generally good but some lack format/constraint details. The server demonstrates solid foundational quality but falls short of A-grade polish.
Parks a feature idea so it does not distract from the locked one-liner. The backlog remembers so the builder does not have to.
Lock the idea and open the build hop. Validates the idea against four gates: problem specificity, ICP specificity, one-liner clarity, and out-of-scope definition.
Advances to the next hop (idea → build → ship → roast → done). Validates that the current hop is complete before advancing.
Delivers candid feedback on the shipped artifact: compares it against the locked one-liner and ICP, ranks dimensions (idea, product, potential), and suggests one concrete next validation move.
Records a shipped URL and summary. Validates that the URL is real (not localhost), deployed, and accessible. Logs the ship to the ship log.
Displays the ship log: a markdown record of every sprint that made it to a URL, with links and summaries.
Output schemas not documented. Tools return text responses but LLMs cannot infer structure of returned data (e.g., sprint_status returns clock line and next action, but format is undocumented). Agents cannot reliably parse or chain results.
Error handling lacks recovery guidance. Tools validate inputs (e.g., URL validation in ship, idea gates in lock_idea) but error responses do not tell LLMs what to do next or how to self-correct. No actionable error messages.
next_hop naming is ambiguous. Does it advance to the next phase, or retrieve the next phase? Verb 'next' is not an action verb (get_, update_, advance_). Should be 'advance_hop' or 'complete_hop' for clarity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Reports the current sprint state: active or done, current hop, time remaining overall and in the current hop, and the next action.
Starts a timeboxed ship sprint (the Habitat one-night format: lock the idea, build, ship, roast). Call with no arguments first: the bunny will tell you what to ask the builder (solo, team, or event). Then call again with mode and details. State lives in the local .habitat/ folder.
check_scope description is vague. 'Parks a feature idea' is domain-specific jargon. LLMs may not understand when to call this vs. other tools. Should explain: 'Temporarily stores a feature idea without locking it into the sprint, preventing scope creep while preserving ideas for later.'
roast tool has no input parameters documented. Description says 'Delivers candid feedback' but does not explain what data it uses (locked idea? shipped URL? ICP?). Implicit dependencies on prior state are not documented.