Vibe Coding Discoverjev-mcp
View on GitHubProof of concept MCP for Typesafe's new Jev AI model
MCP server wrapping TypeSafe's Jev (System One) model as three typed judgment tools: verify claims against evidence, screen text for prompt injection, and rank candidates semantically with no embeddings. Returns calibrated probability distributions in ~150-500ms.
Use Cases
Verify claims against cited evidenceScreen fetched web content for prompt injectionRank candidates by meaning without embeddingsCitation checking in reports and PRsAgent guardrails before reading a pageSemantic find over documents or filesCheap pre-filtering before frontier model callsLow-confidence escalation to humans
Built With
- Language
- JavaScript
- Frameworks
- Model Context Protocol SDK · @typesafe-ai/sdk · Zod · Node.js
Tags
mcp · model-context-protocol · typescript · verification · guardrails · prompt-injection · semantic-search · judgment-tools · typesafe · jev · no-embeddings · confidence-calibration