Bookmarks
Top 10 GitHub Repos
- Kimi Code CLI
- An AI coding agent that runs in your terminal — it can read and edit code, run shell commands, search files, fetch web pages, and choose the next step based on the feedback it receives. Works out of the box with Moonshot AI's Kimi models and can be configured to use other providers.
- MCP TypeScript SDK
- The official TypeScript SDK for Model Context Protocol servers and clients, now on the v2 line (@modelcontextprotocol/server, @modelcontextprotocol/client) implementing the 2026-07-28 MCP spec. The v2 rework addresses long-standing architectural issues, and the docs start with a ten-minute server tutorial.
- Agent-Native
- A framework for building agent-native applications where one defineAction() powers every surface: UI, agent, HTTP, MCP, A2A, and CLI. From BuilderIO, it's a "don't pick between apps or agents" approach that unifies app and agent entry points behind a single schema.
- ego-lite
- A browser built for AI agent automation that shares your logged-in browser state with agents like Codex or Claude Code, letting them run multiple tasks in parallel "Spaces" while you work undisturbed. Zero cost, zero config.
- OmniRoute
- A free MIT-licensed AI gateway exposing one endpoint to 290+ providers (90+ free) and 500+ models, with clients for Claude Code, Codex, Cursor, OpenCode, Cline, and Copilot. Includes quota-aware auto-fallback, RTK+Caveman compression that claims 15–95% token savings, and MCP/A2A support.
- Agent Governance Toolkit
- Microsoft's public-preview toolkit for shipping autonomous agents to production: policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering, with coverage mapped to 10/10 of the OWASP Agentic Top 10.
- Speech To Speech
- A low-latency, fully modular voice-agent pipeline — VAD → STT → LLM → TTS — exposed through an OpenAI Realtime-compatible WebSocket API. Every component is swappable, and the LLM slot accepts any OpenAI-compatible endpoint, from hosted providers to local models via HF Inference Providers.
- T3 Code
- An open "agent harness control surface" that lets you control the agents on your machine — Claude Code, Codex, Cursor, Grok Build, and OpenCode — from iOS/Android, a web app, or an Electron desktop app. It works with the subscriptions and agents you already have set up on your computer.
- OpenSRE
- An open-source framework for building AI SRE agents, including the training and evaluation environment they need to improve. Connects 60+ tools you already run, lets you define custom workflows, and investigates incidents on your own infrastructure — currently in public alpha.
- jCodeMunch MCP
- An MCP server for precise, symbol-level GitHub source retrieval via tree-sitter AST parsing, compatible with Claude Code, Cursor, and any MCP client. Claims 95%+ token-cost reduction on code exploration, with 313B+ tokens already saved.
Honorable Mentions
- andrewyng/aisuite
- Simple unified interface to multiple GenAI providers; the new OpenWorker desktop coworker now lives in its own repo.
- modelcontextprotocol/python-sdk
- Official MCP Python SDK v2, same 2026-07-28 spec support as the TypeScript SDK.
- different-ai/openwork
- Open-source alternative to Claude Cowork for sharing skills, MCPs, and connected services across agents.
- alibaba/open-code-review
- AI code review CLI mixing deterministic pipelines with an LLM agent, battle-tested at Alibaba scale.
- block/buzz
- Self-hostable workspace where humans and agents collaborate in shared rooms on a relay you own.