Frameworks & SDKs tools for Claude Code

by Anthropic - Anthropic's official, curated directory of high-quality Claude Code plugins, installable from within Claude Code

(31.1k ⭐) - Reference agents, skills, and data connectors for the financial-services workflows

(20.6k ⭐) - Open source repository of plugins primarily intended for knowledge workers to use in Claude Cowork & Claude Code

Beads - A memory upgrade for your coding agent

Self-healing browser harness that connects an LLM directly to your real Chrome via CDP. The critical design decision: the agent itself writes missing helpers and domain skills into the harness during execution, so the scaffold improves every run rather than requiring manual updates. At ~1k lines across four core files, it is the clearest published demonstration of an editable harness that learns f

(7.3k ⭐) - The official Python SDK for Claude Code

(5.8k ⭐) - Skills for threat modeling, scanning, triage, patching, plus an autonomous scanning harness you can /customize

by Anthropic - An official AI-powered security-review GitHub Action that uses Claude to analyze pull-request diffs for vulnerabilities

(1.8k ⭐) - A tool for building Claude Code requirements

(1.4k ⭐) - A personal Claude Code Development Kit

(1.2k ⭐) - The ultimate AI development stack, including Claude Code, Task Master, and Curso

Terminal coding agent and Python agent framework built on Pydantic AI, with planning, sub-agents, MCP and optional Docker-sandboxed execution; works with any model, including local ones

Autonomous agent framework that runs unattended on GitHub Actions, triggered by cron schedules or repository events. Dispatches skills to six coding-agent harnesses behind one contract (Claude Code, Grok, Codex, Pi, Vibe, Kimi), defines agent behavior as Markdown skills, persists memory in the git repository, evaluates its own run output to revise underperforming skills, and ships an MCP server ex

Local-first database and workspace for AI agents that keeps structured data, durable knowledge, documents, files, skills, and apps in one searchable place. Agents connect through MCP, OpenAPI, CLI, or coding-agent skills, and material writes can remain reviewable ChangeRequests before becoming canonical

Coordination protocol above Git: agents declare intent, semantic scope, and operation before writing code, deterministic rules raise findings when plans collide, and acceptance is gated on a verification run it executes itself. One Rust binary over local SQLite, with MCP for Claude Code, Codex, and Cursor.

Routes one coding thread across harnesses with quota-aware rotation between subscription profiles, Best-of-N runs, and cross-family review.

(476 ⭐) - TypeScript toolkit for agent-authored video, motion graphics, compositing, and 3D scenes, with native WebGPU rendering and video export

Separate signer that allows, denies or holds each agent tool call for a person before it runs and records every call in an Ed25519-signed, hash-chained log that verifies offline; Python and TypeScript SDKs, framework integrations, coding-agent hooks and an MCP gateway

Shared context, memory, and task coordination as a single Go binary over local SQLite with hybrid keyword and semantic search.

MCP gateway that scopes agent tool access by identity, team, and role, with security policy enforced on each action and an audit trail. Useful as a reference for putting authorization at the tool boundary instead of relying on model prompts to restrict access

Open-source context and control plane for AI agents. Connect company tools once and share permission-aware context across ChatGPT, Claude, Codex, OpenClaw and custom agents, with MCP, scoped identities, approvals and audit

Deploys versioned agents, skills, rules, and governed workflows into the native project paths used by Claude Code, Codex, Copilot, Cursor, and other coding agents.

(206 ⭐) - Configure models, enable tools, stream events, then fetch text, JSON, run details or token stats in one call via .asText() or .allowTools('Read', 'Write')

Multi-agent workflows and complex Agent interactions, both via YAML manifest and programmatic usage. Pydantic-AI and LiteLLM backends with human-in-the-loop integration

(177 ⭐) - Fast, intelligent quality checks for different project types

(170 ⭐) - A high-performance Rust implementation of an OpenAI-compatible API gateway for Claude Code CLI

Enterprise-grade orchestration framework that coordinates multiple Gemini CLI instances for complex development tasks, based on battle-tested Claude Code Flow patterns

Visual prompt builder: decompose prompts into semantic blocks and compile them into Claude-optimized XML

Delegates a task to DeepSeek, Codex, or Claude from inside your current Claude Code or Codex session, returning the result automatically.

Self-hosted AI agent platform (Apache-2.0): one chat routes work to a team of agents on any model or on your own Claude Code, Codex and GitHub Copilot CLI sessions, with Playbooks, multi-agent Missions, a task board, RAG and a Knowledge Graph

Persistent cross-session memory MCP server for 13+ AI coding tools. SQLite-backed state survives context resets and keeps Claude Code, Cursor, Gemini CLI, Codex, Windsurf, Amp, Kiro, VS Code Copilot, Trae, and more in sync. Install: npm install -g @egchq/egc

The team hivemind: persistent context recorded once and recalled across agents, machines, and teammates, as a Go CLI.

CLI-first control plane that turns one-off coding-agent calls into reviewable Markdown workflows spanning Claude Code, Codex, Gemini CLI, or Copilot CLI. Explicit artifact handoffs keep execution inspectable on disk, while validated completed nodes let failed workflows resume instead of starting over.

Watches your real local sessions and surfaces them in a live triage inbox, keeping a searchable record with line-level agent attribution. Claude Code, Codex, Cursor, Gemini.

Open standard for agent-to-agent messaging with Ed25519-signed envelopes, federation across providers, and delivery that must be proven rather than assumed. Reference server, TypeScript SDK, and a Claude Code plugin.

Takes what a coding agent produces and commits each turn to git, recording the prompt, model, and that turn's token cost in the commit message, with the agent confined to its own worktree. Claude Code, Codex, and OpenCode.

Durable AI agent runtime with MCP client/server and A2A support: graph-based workflows, durable execution and multi-agent coordination, authored in Python

Python SDK for Gemini CLI; API-compatible with Claude Code SDK

MIT-licensed context compactor for coding-agent transcripts (Claude Code, Codex CLI): moves spent tool outputs to a local content-addressed stash the agent can recall by hash, keeps user and assistant messages byte for byte, and writes a JSON report per pass; CLI, MCP server and Claude Code plugin, offline

Declares skills in YAML and pins exact revisions in a lockfile so installs are reproducible across Claude Code and Codex.

Python CLI that compiles contract-based YAML runbooks into SKILL.md files for Claude Code and Codex agents. Define multi-step agent workflows with loops, branching, parallelism, and file-based state passing — write the runbook once, generate executable SKILL.md with one command. pip install git+https://github.com/KnoxOps/agent-runbook.git

Runtime trust scanner for agent skills and tool use: a tested pattern library across 100+ attack categories (prompt injection, tool poisoning, MCP attacks, skill compromise; current counts in the README) plus a mechanism layer, shipped as a pip package, GitHub Action, and free web scanner with a published, reproducible precision/recall benchmark

Self-evolving AI agent framework with 8 specialized agents, SEED protocol (8-phase recursive self-improvement), knowledge transfer, and MMORPG progression. Supports 15 models across 7 providers. github | website

Persistent AI memory for coding assistants. Auto-captures decisions, patterns, and context across sessions. VSCode extension + CLI + MCP server. MIT licensed

Self-hosted Telegram agent that breaks down short videos locally (Whisper transcript, OCR, extracted commands, repos and links) and plans and builds what they show with open models, asking on the phone before each shell command

TypeScript SDK with defineHook(), typed event payloads for all 5 hook events, response builders, and unit-testable .handle() method. Zero dependencies

MIT Python SDK that traces LLM calls, tool calls and agent steps as OpenTelemetry GenAI spans, with auto-instrumentation for OpenAI, Anthropic, LangChain, LlamaIndex, LiteLLM and MCP, and exports over OTLP to Rius or any OTLP-compatible backend

Hooks plugin for Claude Code and Codex that quotes the user's own past corrections back to the agent before it answers, reports what the user corrects most, and gives the orchestrator model-selection guidance for subagent calls. Smoke-tested on macOS: Codex 0.162.1 and Claude Code 2.1.311 (headless, correction capture and next-prompt recall only); interactive Claude Code use is experimental. MIT,

Read-only shell diagnostic for scheduled AI-agent chains; checks launchd, cron and systemd state and logs for the ways an unattended agent stops silently

Free GitHub Action that scores public repositories for Codex, Claude Code, Copilot, and Cursor readiness—without cloning or executing code

Secrets stay on a paired device and are lent to the agent's machine into daemon memory only, served over FUSE or loopback WebDAV; the daemon zero-wipes them when the pairing drops or a grace window expires