Devotd Github
NewAI agents

varshavia/repomagpie v0.2.1 — Tool choice records for agents

RepoMagpie is a local tool that lets a coding agent keep records of the user's evaluations of open-source tools so it can refer back to them when choosing a tool.

Repositoryvarshavia/repomagpie ↗

RepoMagpie is a local tool that lets a coding agent keep records of the user's evaluations of open-source tools so it can refer back to them when choosing a tool.

GitHub description: RepoMagpie: your coding agent remembers the repos you’ve tried, and what you thought of them.

stars
★ 9
Forks
0
Language
TypeScript
License
MIT
Latest release
v0.2.1
Read original source ↗View on GitHub ↗

RepoMagpie (varshavia/repomagpie) is a local record tool that lets a coding agent refer back to the user's own evaluations of the open-source repositories and tools the user has tried. The latest release is v0.2.1, but the provided record does not state what changed in this version. The repository is a TypeScript project published on October 1, 2026.

Evaluations are kept as markdown files and are designed to be checked through search and recommendation before a tool is chosen. The user evaluates libraries, command-line tools, MCP servers and agent skills directly, and the agent recommends or avoids them by separating items already used from items the user does not want to use. It is also provided as an MCP server, so agents can reference the records. This is a setup that does not leave tool choice to the agent's generic recommendations alone, but tries to reflect the user's past evaluations. The license is MIT. The README explains integration with seven coding agents and the requirement of Node.js 22.12 or later, and the installation notes include checking and backing up before configuration changes, along with a run check command.

With only the provided material, however, neither the changes in v0.2.1 nor the quality of the recommendations can be confirmed. The names of the seven agents and any real adoption cases are not given either. The purpose of this project, recording personal evaluations and feeding them into tool choice, is therefore clear, but how useful the recommendations really are is hard to judge from the current material.