twiceshy
A self-hosted MCP server that retrieves git-backed engineering experience records (issues, fixes) to inform LLM coding agents, preventing repeated mistakes with a relevance floor.
README
twiceshy
Once bitten, twice shy.
A self-hosted service that feeds hard-won engineering experience — issues, dead-ends, root causes, validated fixes — to LLM coding agents at decision time, so they stop repeating known mistakes on autopilot. Records are git-backed and provenance-aware; automatic retrieval applies a relevance floor so a weak match can return nothing.
Hosted alpha
The hosted instance runs at twiceshy.app — self-serve
tokens, remote MCP endpoint https://api.twiceshy.app/. External
contributions require a signed CLA before merge
(ADR-0002).
How it works (the locked architecture)
- Source of truth: git-backed markdown experience records (YAML frontmatter
- narrative) — format in docs/SCHEMA.md. The corpus is a
separate versioned data product (
twiceshy-corpus, ADR-0021); the engine ships a small frozen fixture (internal/testcorpus/) for tests.
- narrative) — format in docs/SCHEMA.md. The corpus is a
separate versioned data product (
- Index: one derived, always-rebuildable SQLite file (FTS5 plus cached, pure-Go dense vectors for pull retrieval).
- Retrieval: fingerprint-exact → BM25 → dense (RRF), stack-fingerprint filtered, hard cap k≤3 with a relevance floor — below it, nothing is injected.
- Channels: push (Claude Code hooks → trap cards), pull (MCP tools over streamable HTTP), index (generated one-liners).
- Trust: agent-proposed records are quarantined; promotion = sandbox fail-to-pass validation + human PR review. A new record IS a pull request.
- Doctors: background jobs that dedup, staleness-check, re-execute repros (CI for memories), decay, and abstract.
Full rationale: docs/research/EXPERIENCE_SERVICE_RESEARCH.md and docs/adr/ADR-0001-architecture.md.
Status
The core service is available as a hosted alpha. For the canonical snapshot of what ships now and the current roadmap, see docs/NEXT_FEATURES.md; deployment and operating details live in docs/DEPLOY-public-alpha.md.
Development
make ci # lint + race tests + coverage floor — what CI runs
See docs/CONVENTIONS.md and docs/CONTEXT.md first.
License
AGPL-3.0-only. Contribution and corpus licensing: docs/adr/ADR-0002-licensing-strategy.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.