Your AI knows the code. It doesn't know your playbook.

Mimir ships your team's playbooks to Cursor, Windsurf and Claude Code — so the AI stops guessing and starts working the way you work.

Task your AI with confidence

Bring your knowledge library to the IDE you already love.

One MCP server, every AI IDE. Your rules, skills and workflows follow you from Cursor to Claude Code — no copy-paste, no re-onboarding.

See how it plugs in

Catch what your team knows before it walks out the door.

Turn the unwritten rules of "how we do things here" into versioned playbooks the whole family can run — and the AI can read.

Preserve what you know

Plan the sprint once. The AI already knows the plan.

Playbook activities line up with your GitHub or Jira tickets, and Mimir's MCP layer gives the AI both sides in the same breath. Less briefing, more shipping.

Run a sprint with AI

When a project drifts, bring it back to the book.

Compare what actually happened against the playbook, spot the drift, and ship the lesson back as a versioned improvement proposal. Then it won't happen again.

See audit & recovery

Your AI finally knows what "done" looks like here (and it stops re-learning it every Monday)

You already noticed it: the assistant is brilliant at the language, clueless about your standards. Test framework, layering rules, how PRs get reviewed, the one thing you never ship without — it guesses, and it guesses differently every session. Mimir hands it a versioned playbook, the same one your team uses, so the guessing stops.

One MCP, every IDE (Cursor, Windsurf, Claude Code — your call)

Mimir runs as a local FOB container next to your editor. One MCP server exposes your playbooks, rules, skills and artifacts to any AI assistant that speaks MCP. Switch models this quarter, switch IDEs next quarter — your knowledge library doesn't move.

A playbook is a book, not a ticket system (tracking stays where your team already tracks)

Mimir doesn't try to replace GitHub Issues or Jira. Playbooks are static, versioned reference material — the "how we do it here." Your existing trackers own the "is it done yet." The MCP layer lets the AI read both, so you get one coherent picture without duplicate systems of record.

Co-author your playbook with the AI that runs it (export, edit, review, import — versions all the way down)

Export a workflow into your repo as Markdown files. Let the AI propose additions, re-orderings, better guidance. Review the diff in your IDE like any other PR. Import back, and Mimir either applies the changes to your draft or files them as a Playbook Improvement Proposal for the released version. Every edit is traceable; nothing leaks into main without a rationale.

Ship your methodology like software (families adopt, versions evolve, approvals gate the rest)

Publish a playbook to your Homebase family — your team, your consultancy, your community — and everyone with access can sync it into their local FOB. Admins approve, versions ship, improvements flow back as proposals. The "way we do things here" finally has a release train.

Playbook rules land as `.mdc` files in your IDE (the AI can't miss them even if it tries)

Every playbook carries rules — the non-negotiables. Mimir exports them as IDE rule files (rules/*.mdc) with alwaysApply front-matter, so Cursor, Windsurf and Claude Code enforce them on every turn. "Use pytest", "no any in new code", "never write to prod without a migration" — on-rails, automatically, for every teammate.

Ship your first playbook this afternoon

Mimir is a local container. Pull it, wire it to your IDE's MCP settings, and you can have a draft playbook in your assistant's hands in under an hour — no account, no credit card, no Homebase until you want one. When you're ready to share with your team, flip on the Homebase connection and publish.

Get the Docker quick-start See the user journey

Free local FOB · Optional Homebase for teams · Open MCP so you're never locked in.