Jev in production › Agent tool and action selection

toolJev

Jev replaces LLM-based tool selection in MCP agents across 612 tools; author reports 98.8% accuracy on 400 tickets (author).

Open on GitHub ↗

98.8%measured against a baseline, as published by the source
Use
Agent tool and action selection
Industry
Developer tools
Form
Open-source tool
Stage
Beta
Plugs into
MCP
Listed
2026-09-29
Found via
github
Repository
VishiATChoudhary/toolJev
Stars
3
Forks
0
Last push
2026-09-28
Language
Python
License
MIT
toolJev screenshot
docs/results_card.png in the VishiATChoudhary/toolJev README, MIT; shown from GitHub.

The README opens with

Code Mode for MCP, where the sub-model is a decision model, not an LLM.

Your agent doesn't need an LLM for every decision. 612 tools: 77% fewer tokens. 400 tickets: 5.3x cheaper, 98.8% accurate.

I benchmarked toolJev with hosted Jev on MCPToolBench++, LiveMCPBench, When2Call and live Claude Haiku agents. Several results went against my first design, and the design changed to match.

Question Result (hosted Jev) Takeaway --------- Does it pick the right tool? Right tool ranked first: 83% on MCPToolBench++, 54% on LiveMCPBench, 99% on When2Call. Retrieval alone: 73%, 40%, 92%. Jev alone, with no retrieval: 18% Retrieval shortlists, Jev picks Does it know when no tool fits? AUROC 0.94 on When2Call near-misses (embeddings 0.74), 0.72 on...

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for agent tool and action selection