What does MCP Router do?

It discovers, catalogs, classifies and deduplicates every MCP server and tool you have, then exposes each AI agent only the small, relevant, authorized subset it needs for the task instead of flooding the context window with the whole catalog.

What does Jev-based mean here?

Routing decisions are System One decisions — typed Choice, Score or Noul questions answered with calibrated probabilities in a single forward pass rather than generated text — and the decision engine is pluggable across edge, hosted and Azure OpenAI backends.

What is the recommended decision engine?

Laya, a 421M open-weight, Jev-compatible decision model under Apache 2.0, opt-in through the inference extra. Out of the box the router runs a zero-ML deterministic backend, and any model failure or timeout degrades to deterministic retrieval ranking flagged fallback_used.

How does retrieval work?

Embeddings live in PostgreSQL with pgvector — hash projection by default, BGE-small with the inference extra — fused with Postgres full-text keyword search using reciprocal-rank fusion, so the stack also runs with no torch installed.

Can model scores widen what an agent can access?

No: a deterministic, deny-by-default policy engine and execution manager gate everything, and the README states model scores are relevance data that can never widen access. Exposure budgets can be lowered by a request but never raised, and the route cache re-checks authorization on every hit.

How are agent skills handled?

Skills are routed like tools through one pipeline and one policy engine: sources can be a local directory or an https git repo, each skill is classed read, write, execute or unknown with unknown treated as execute, and policy rules use resourceKind skill under a separate maxSkills budget.

What analytics does it provide?

A funnel following each routing decision from surfaced to selected to succeeded, with execution-to-decision attribution, position bias, context-token savings versus the agent's full authorized catalog, per-agent profiles, wasted-exposure and staleness suggestions, daily rollups and Prometheus totals.

Jev directory

CommunityJev Agents & MCP

MCP Router

An open-source Jev-based MCP router: it discovers, catalogs, classifies and deduplicates every MCP server and tool you have, then exposes each agent only the small, authorized subset it needs for the task. Routing decisions are System One Choice/Score/Noul questions, scored by a pluggable engine at the edge or hosted.

See highlights Open github.com

Stars: 0
Language: Python
License: APACHE-2.0
Reading time: 3 min
Published by: Community
Author: amantux
Last verified: Oct 2026

amantux/mcp-router

User repository on GitHub

View on GitHub

Open-source Jev-based MCP router with edge capacity: System One (choice/score/noul) decision routing over Laya, hosted Jev, or Azure OpenAI; pgvector retrieval; deny-by-default policy gateway

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw mcp routing policy

Works forJev for Routing Jev for Tool calling Jev for Decision gating

Used inJev for Software & DevTools

Languages: Python
Platforms: MCP PostgreSQL API
Capabilities: Open sourceSelf-hostedRuns locally

Highlights

The decision engine is pluggable: run Laya, a 421M open-weight Jev-compatible model (Apache 2.0), at the edge, use a hosted endpoint, or Azure OpenAI.
Retrieval uses Postgres with pgvector plus full-text keyword search fused by reciprocal-rank fusion; the stack runs torch-free via hash embeddings.
A deterministic, deny-by-default policy engine and execution manager: model scores are relevance data and can never widen access.
Exposure budgets cap exposed tools and distinct servers per request, per agent and globally, and a request can lower a budget but never raise it.
Analytics follow every decision surfaced to selected to succeeded, with per-agent profiles and Prometheus totals; cache hits count as real traffic.

Watch out

Grounded in the truncated README and GitHub metadata reviewed on 2026-10-15; the excerpt cuts off mid-sentence in the Skills section so the skills workflow is only partially documented, no benchmarks are published, retrieval assumes Postgres with pgvector, and hosted or Azure backends need their own URL and key.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.

FacebookClinicnotes

X: OpenRouter Decision Model Rankings

X: herdr + jev, an agent router

X: A group chat bot with decision room routing

From the Jev guides

Original write-ups on what this entry is for.

MCP tool routing: Jev in front of the tool call

Oct 16, 2026

Tool menus can outgrow the context that reads them. The catalog's Jev-based routers and gates narrow the list: benchmarked tool search, per-tool scoring, fail-closed approvals, and WebMCP in the browser.

Jev Agents & MCP#mcp#tool-routing#agents

Read article

Approval policies for coding agents: hooks, guards, and gateways

Oct 16, 2026

How the Jev guard family turns coding-agent permissions into auto-approve, deny, or ask: hooks that enforce, guards that fail safe, and gateways that gate generation.

Jev Agents & MCP#coding-agents#guardrails#hooks