An open-source MCP server that gives your AI structured memory, epistemic discipline, and the ability to build on what came before.
Your AI forgets everything between sessions. It guesses instead of checking. It can't tell a brainstorm from a decision. oddkit fixes that.
oddkit reads markdown files from a GitHub repository — decisions, constraints, learnings, governance — and makes them available to your AI through structured tools. It works with any AI tool that supports MCP: Claude, ChatGPT, Gemini, Cursor, Claude Code, Lovable, Replit, ElevenLabs voice agents, and more.
Knowledge base repo: klappy/klappy.dev — the content oddkit reads from
oddkit is a remote MCP server. You don't install anything — you point your AI tool at a URL.
You can optionally add ?consumer=yourname to the URL to identify yourself on the public transparency leaderboard. oddkit tracks which tools are used and how often, but never what you search for or what documents contain. Replace yourname with your name, a handle, or a project name — or leave it off to stay anonymous.
Settings → Connectors → Add Custom Integration:
- Name:
oddkit - URL:
https://oddkit.klappy.dev/mcp?consumer=yourname
Settings → Developer Mode → Create App → add MCP server URL:
https://oddkit.klappy.dev/mcp?consumer=yourname
Add to your .mcp.json:
{
"mcpServers": {
"oddkit": {
"type": "http",
"url": "https://oddkit.klappy.dev/mcp?consumer=yourname"
}
}
}Or in Claude Code: claude mcp add --transport http oddkit https://oddkit.klappy.dev/mcp?consumer=yourname
Any tool that supports MCP can connect. Look for "MCP server" or "custom integration" in your tool's settings and provide the URL:
https://oddkit.klappy.dev/mcp?consumer=yourname
Once connected, your AI gets access to these tools:
| Tool | What It Does |
|---|---|
| orient | Assess a situation, surface unresolved questions, identify which mode you're in (exploring, planning, executing) |
| search | Find relevant documents, constraints, and prior decisions by topic |
| get | Fetch a specific document by URI |
| challenge | Pressure-test a claim, assumption, or proposal against existing constraints |
| gate | Check readiness before transitioning between phases |
| encode | Structure a decision, insight, or boundary as a durable record |
| preflight | Pre-implementation check — surfaces constraints, definition of done, and pitfalls |
| validate | Verify completion claims against required artifacts |
| catalog | List available documentation with filtering and sorting |
After connecting, say "use oddkit" or "ask oddkit" to invoke it:
- "Use oddkit to orient me on whether I should [decision you're facing]"
- "Ask oddkit to challenge my assumption that [something you believe]"
- "Use oddkit to encode this decision: we chose [X] because [Y]"
- "[paste meeting notes] Use oddkit to encode the key decisions from this meeting"
To make oddkit proactive — so the AI uses these tools automatically instead of waiting for you to ask — add a bootstrap prompt to your project instructions. See the full bootstrap guide or start with the essentials in Getting Started with ODD and oddkit.
By default, oddkit reads from klappy.dev. You can point it at any GitHub repo using the canon_url parameter:
canon_url: "https://raw.githubusercontent.com/YOUR_ORG/YOUR_REPO/main"
oddkit reads markdown files with YAML frontmatter. Start with a few files — decisions, constraints, learnings — and grow from there. No schema required.
oddkit is a Cloudflare Worker that:
- Fetches markdown files from a GitHub repository (zip download, cached)
- Indexes them with BM25 full-text search
- Parses YAML frontmatter for metadata, filtering, and sorting
- Exposes structured tools via the MCP protocol
It's stateless, serverless, and framework-agnostic. The knowledge base is your repo. oddkit just makes it searchable and structured.
cd workers
npm install
npm run dev # Local development
npm run deploy # Deploy to CloudflareBranches:
main→ staging previewprod→ production (oddkit.klappy.dev)
Promote staging to production: ./scripts/promote.sh
- Getting Started with ODD and oddkit — five-minute quickstart with bootstrap instructions
- The Journey from AI Tasks to AI-Augmented Workflows — the four-step progression
- From Passive to Proactive — the story behind oddkit's proactive design
- klappy.dev repo — the knowledge base oddkit reads from
MIT