03 / AI REPO RADAR
Compiled daily by an AI agent: rising GitHub repos for MCP, agent frameworks, coding tools, evals and memory.
This list is compiled automatically every day by an agent, with automated checks. Report mistakes by email.
4,499 starsLicense: Apache-2.0
A Microsoft plugin that learns from earlier agent sessions in a repository to cut cost: it remembers repo facts, stops loading unused tools and redirects turns that get stuck.
Why it's useful
Helps Claude Code, Codex and Copilot CLI users cut repeated token spend without modifying the agent; every action is reversible.
107,971 starsLicense: Apache-2.0
Persistent memory for coding agents: it records what happens in a session, compresses it with AI and injects relevant context into later sessions, with MCP tools for searching it back.
Why it's useful
Lets an agent remember past decisions and fixes instead of making you re-explain them every session.
111,184 starsLicense: MIT
Open source app for running several coding agents side by side, each in its own git worktree and tracked in one place, with desktop and mobile companions.
Why it's useful
Lets you compare several agents on the same task without juggling terminals and branches by hand.
23,548 starsLicense: Apache-2.0
Security scanner for AI agent skills: finds prompt injection, data exfiltration and supply-chain risks before install, with an MCP server mode.
Why it's useful
Adds a checkpoint before an agent installs third-party skills or MCP servers.
50,344 starsLicense: Apache-2.0
An agent-native learning workspace: chat, problem solving, quizzes, research and visualization share one runtime, with several RAG engines and layered memory.
Why it's useful
A reference for agent builders on merging many modes into one runtime, plugging in several retrieval backends and letting an agent consult other coding agents.
53,775 stars+2,072 / 7dLicense: MIT
Vectorize's agent memory system: stores facts, consolidates them into evidence-backed beliefs and mental models so agents learn over time instead of merely replaying chat history.
Why it's useful
Adds long-term memory to an agent in a few lines (LLM wrapper) or via MCP, with retain, recall and reflect.
71,541 stars+5,674 / 7dLicense: Apache-2.0
Open-source framework that turns HTML, CSS and seekable animation into deterministic MP4 video, with a skill set that lets coding agents plan, write, preview and render videos.
Why it's useful
Lets an agent make video from HTML it already writes well, with repeatable renders in CI.
195,040 stars+2,589 / 7dLicense: AGPL-3.0
Open-source API to search, scrape and interact with the web, returning markdown or structured JSON for agents; connects through a skill, CLI, MCP and several SDKs.
Why it's useful
Gives agents clean data from web pages, PDFs and DOCX, using fewer tokens than raw HTML.
20,274 stars+5,278 / 7dLicense: MIT
A desktop app with a local gateway that sets the model for each coding agent (Claude Code, Codex, Gemini CLI...) from one place and fails over to another account when a quota runs out.
Why it's useful
Run several agents and providers without hand-editing each agent's config file.
373,627 stars+68,203 / 7dLicense: MIT
An MCP server with a CLI that lets agents reverse engineer software: native binaries, Electron/JavaScript, .NET, APKs and websites, returning findings with evidence and limitations.
Why it's useful
Lets an agent inspect an app without source code and explain how a feature works.
77,575 starsLicense: Apache-2.0
A local context-compression layer: shrinks tool outputs, logs, RAG chunks and files before they reach the LLM, usable as a library, a proxy or an MCP server.
Why it's useful
Long agent sessions with heavy tool output burn tokens; Headroom plugs in via a proxy or agent wrapper, no code changes, and originals stay retrievable.
106,209 starsLicense: MIT
An open-source app (Node.js server plus React UI) that orchestrates a team of AI agents like an organization: task assignment, org chart, budgets, approvals and run history.
Why it's useful
When juggling many agents in parallel, Paperclip gathers tasks, costs and permissions in one place and works with existing agents like Claude Code or Codex.
100,696 starsLicense: MIT
A capability layer that gives AI agents internet access: one CLI that picks, installs and health-checks the tools for reading the web, YouTube, GitHub, Twitter/X, Reddit, Bilibili and RSS.
Why it's useful
Agents stop getting stuck at the 'look it up online' step: each platform has a primary and fallback backend, and the doctor command shows which one is live.
12,458 starsLicense: Apache-2.0
Local-first search engine and MCP server in one Rust binary: indexes code, docs and logs so agents query the exact passage instead of reading files one by one.
Why it's useful
Cuts context tokens when an agent works on large codebases or code the model has not memorised.
30,697 starsLicense: Apache-2.0
Open-source diagnostic that independently audits AI agents: runs checks on permissions, prompt injection and data leakage, then returns a letter grade with JSON and Markdown reports.
Why it's useful
Gives agent builders a repeatable check on whether an agent stays within its assigned role and permissions.