Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-skill-router

Rust MCP server that retrieves agent-skill metadata with SQLite FTS5 and ranks candidates with one typed Score each from Jev.

MCP serverbeta
View on GitHub ↗

Overview

Jev Skill Router recommends third-party agent skills for a project. Offline Python tooling indexes skill metadata from GitHub into SQLite with FTS5/BM25; a native Rust stdio MCP server then searches that index and sends one comparable typed Score per candidate, in batches of up to 128 questions, to a configured decision provider. It never installs or executes skills.

It is for agent users who want help discovering skills from a large external catalog. Providers are configured in a JSON profile file: the default profile calls Jev at TypeSafe's systemone endpoint, and a Perplexity pplx-decider profile is also shipped. Any endpoint that accepts the same model/state/questions request and returns typed Scores can be added without code changes. There is no automatic fallback between providers. Retrieval needs no model key; ranking does.

What the decision model decides

TypeQuestionAnswers
ScoreHow well does this candidate skill fit the project or request?(inferred)—

How it works

  1. Python indexer pulls skill metadata from GitHub into SQLite FTS5
  2. MCP search retrieves and deduplicates a BM25 shortlist, applying exclusion policy first
  3. Rust service sends one Score per candidate (max 128 per batch) to the selected provider
  4. Ranked recommendations are returned; nothing is installed or executed

Strengths

  • Provider-agnostic: Jev and pplx-decider profiles, plus any compatible typed-decisions endpoint
  • Runtime is a single native binary with read-only SQLite and no subprocesses
  • Retrieval works without any model key
  • Has a CI verify workflow per the README badge

Caveats

  • Setup needs Rust, Python with FTS5 and an authenticated GitHub CLI
  • Provider changes require a process restart
  • README notes dated receipts are historical evidence, not proof of a current deployment

Getting started

Clone, install requirements-offline.txt, run `python3 -B router.py index --workers 16`, `cargo build --release --locked`, then start the stdio server with SKILL_ROUTER_DB and SKILL_ROUTER_PROVIDERS_FILE set.

At a glance

Type
MCP server
Status
beta
Language
Rust
Runtime
MCP server
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does the decision model do here?

It scores each retrieved candidate skill so the shortlist can be ranked by fit.

Can I use something other than Jev?

Yes. A Perplexity decider profile ships, and any endpoint with the same typed Score request and response shape can be added in the providers file.

Does it install skills for me?

No. It only recommends; it never installs or executes skills.

Related projects

  • matchcn

    MCP server and semantic index that tags shadcn-format components once and matches a plain-language brief to them deterministically.

    MCP serverbetaTypeScript
  • apple-rag-mcp

    MCP server over Apple developer docs and WWDC transcripts that uses hybrid search, then Jev relevance scores to rank results.

    MCP serverproductionTypeScript
  • hippo-memory

    Local long-term memory for coding agents with an optional Jev reranker that orders recalled memories.

    ProjectproductionTypeScript
  • BGTS Context Engine

    Deterministic code-graph context for coding agents on Postgres, AGE and pgvector, with optional Jev or local re-selection of symbols.

    MCP serverbetaPython
  • ego-jev

    Agent skill where Ego Lite drives the browser and Jev resolves semantic choices: targets, filtering, ranking and evidence checks.

    ProjectbetaMarkdown + TypeScript
  • Jev Sift

    MCP tool that asks Jev which files, web pages or snippets are relevant to a query, so the agent reads selectively.

    MCP serverbetaTypeScript