Harnesses & Runtime Infra tools for MCP

The all-in-one Desktop & Docker AI application with built-in RAG, AI agents, No-code agent builder, MCP compatibility, and more

Self-hosted multi-model chat. All major providers

Open-source, community-driven terminal coding agent harness written in Rust with a model-agnostic runtime, OS-level sandbox (Seatbelt/Landlock/seccomp/bwrap), resumable fleets via an append-only ledger, and a /model command that lets you switch providers mid-task. A strong example of treating the harness — not the model — as the product surface, with clear separation between interactive TUI, headl

NEW (Apr 2026) Agentic focus. 1M ctx. Repo-level coding. MCP-native

SpaceXAI's open-source terminal coding agent harness: a Rust-based fullscreen TUI with an extensible tool runtime, MCP/skills/hooks support, and headless/embedded modes via ACP. A useful first-party counterpoint to Claude Code and Codex CLI for studying how a new model provider structures the loop, checkpoints workspace state, and exposes the same harness through interactive and scripted surfaces

A compact, inspectable open-source agent harness from HKUDS (April 2026) featuring a built-in personal agent (ohmo), auto-compaction with session preservation, MCP HTTP transport, and multimodal gateway support. Excellent reference for understanding how a small, modular harness can support multi-day sessions without manual context management

(11.6k ⭐) - The power of Claude Code + Gemini / OpenAI / Grok / OpenRouter / Ollama / Custom Model working as one

High-performance AI gateway written in Go. Unifies 23+ providers behind one OpenAI-compatible endpoint with automatic failover, adaptive load balancing, semantic caching, virtual keys and budgets, MCP tool calling, and native Prometheus observability

Moonshot AI's official open-source coding agent harness (May 2026): a single-binary, millisecond-startup TUI with built-in coder/explore/plan subagents in isolated contexts, lifecycle hooks, conversational MCP configuration (/mcp-config instead of hand-edited JSON), and first-class video input as a tool surface. Speaking ACP out of the box (Zed, JetBrains) makes it a useful first-party reference f

TrueFoundry's open-source agent harness runtime: runs the execution loop, MCP tools, skills, sandboxing, approvals, context management, and session state, exposing it through a chat UI, HTTP API, and embeddable UI SDK. A concrete reference for turning a model-plus-tools stack into a production-grade agent without building the loop from scratch

All-in-one agent sandbox combining browser, shell, filesystem, MCP servers, and VSCode Server in a single Docker container. Native MCP support exposes sandbox capabilities to LLMs via the standard protocol, and files downloaded in the browser are instantly accessible in terminal and VSCode. Optimized startup (4–8s depending on config) with Claude Skills mounting support. The fastest path to a full

(5.2k ⭐) - Self-hosted MCP server that gives ChatGPT, Claude, and other clients access to approved local projects for file edits, code search, and shell commands

Context runtime and MCP server that reduces AI coding agent token costs via session caching, AST-aware compression, and shell output patterns. Website

A CLI host application that enables Large Language Models (LLMs) to interact with external tools through the Model Context Protocol (MCP)

by dyoshikawa - A Node.js CLI tool that automatically generates configs (rules, ignore files, MCP servers, commands, and subagents) for various AI coding agents. Rulesync can convert configs between Claude Code and other AI agents in both directions

Run Claude Code as a one-shot MCP server -- an agent in your agent. Permissions bypassed automatically. By Peter Steinberger. 1,100+ stars

Persistent context and a code knowledge graph for coding agents, with a Codex CLI launcher

DeepSeek-V4 terminal coding agent harness that treats prefix-cache economics as a first-class design constraint: byte-stable system prompts, cache-reusing forks for reflection and memory, and a self-verifying cross-session memory layer keep real SWE-bench-style tasks at ~95.8% cache hit and ~30× cheaper than Claude Opus. A concrete demonstration that harness-level cache engineering can substitute

Turns production traces into specialised models you own: OpenTelemetry trace ingest, datasets, evals, LoRA fine-tuning of open-weight models and OpenAI-compatible serving. Self-hostable

Scaffold factory that turns any repo into a branded agent harness with its own npx CLI, MCP server, scoped memory, governance policy, and Darwin Mode self-evolution. The clearest open-source embodiment of "the model is replaceable, the harness is the product" — it generates the owned scaffolding rather than prescribing a fixed framework

Linter for AI agent configurations. Validates GEMINI.md, SKILL.md, hooks, MCP, and more with 156 rules, auto-fix, and LSP server for real-time editor diagnostics

One endpoint for every LLM vendor. Any SDK, any model, any tool. One Rust binary

Durable execution layer for AI agents. Checkpoints, replay, resume, and observability primitives that make agent workflows persistent and replayable — no graph DSL required

One API for 25+ LLMs, OpenAI, Anthropic, Bedrock, Azure. Caching, guardrails & cost controls. Go-native LiteLLM & Kong AI Gateway alternative

(245 ⭐) - Connect Claude Code with Google's Gemini AI for powerful AI collaboration

(232 ⭐) - A lightweight integration that brings Google's Gemini AI capabilities to Claude Code through MCP (Model Context Protocol)

by Mykola Maksymenko - Local, multi-provider agent communication tool. Adds hooks and a skill to exchange questions and reviews, and saves decisions, verification results and handoffs for later sessions. Each session keeps its own client, model, conversation and permissions

OpenAI Responses API-compatible server. Run Codex with local models

(178 ⭐) - An MCP server for Claude Code conversation history

Self-hosted platform for building, sharing and governing AI agents across a company, with skills, document search, MCP tools, schedules, budgets, approvals and run history. Built on Pydantic AI

Server application that extends the Google Gemini CLI with MCP toolkit and OpenAI-compatible API bridge

by S.F. Kislev - A tiny bridge that lets Claude Code drive desktop software — Photoshop, Premiere, Blender, Unity, InDesign, Office, 13 apps total — by writing one-time scripts against each app's own automation runtime (COM / AppleScript / CEP), instead of MCP servers or fragile screenshot-based computer use. A genuinely novel approach that unlocks the apps' full scripting surfaces (InDesign alone

Optimize agentic capabilities with agents. Reflective optimizer for AI agent prompts, tool code, MCP tool surfaces, and skill packages; learning from evaluation traces, val-only significance gate with a sealed test split scored once. Adapters for τ²-bench, SWE-bench, SkillsBench. Zero runtime deps, Apache-2.0

Sync Skills, MCP servers, Agents & Commands across Claude Code, Cursor, OpenCode, Codex, and Gemini CLI with automatic format conversion (JSON ↔ TOML ↔ JSONC)

Self-hosted control plane for deploying and operating OpenClaw and Hermes agent fleets on Docker and Kubernetes, with lifecycle management, observability, cost tracking, secrets, schedules, and REST, CLI, and MCP interfaces

Typed Python SDK for the Codex CLI (sync/async, streaming JSONL events, structured items)

Allows the Gemini CLI to send commands to itself, run very long commands in the background and be woken, sleep for a specified amount of time, watch logs and receive notifications, etc by leveraging tmux. Reduces usage and improves reliability by avoiding loop detection. After installing the extension you must lauch gemini-cli into a tmux session named "gemini-cli"

Open-source, self-hosted AI gateway and control plane: OpenAI-compatible LLM routing, RAG, an MCP hub, guardrails and agent tracing for multi-tenant teams

TypeScript CLI that syncs agents, MCP configs, skills, commands, and workflow assets across Gemini CLI, Claude Code, Codex, OpenCode/OpenClaw, and GitHub Copilot workflows

Local-only TUI + web dashboard that tails Gemini CLI sessions alongside Claude Code, Codex, Cursor, Hermes, and OpenClaw on one unified timeline. Parses tokens, tools, and per-turn cost from each Gemini CLI session (gemini-2.5-pro / flash rates), plus context compaction visualizer, MAD z-score anomaly detection, MCP server mode, and OpenTelemetry exporter. No cloud, no telemetry. macOS + Linux. MI

Self-hosted Go Agent Gateway (Apache-2.0) — OpenAI-compatible LLM proxy + MCP aggregation with per-consumer auth/audit

Self-hosted AI agent platform with multi-provider LLM support (Claude, OpenAI, Gemini, Ollama), MCP-native tool integration, visual workflow automation (Chain Engine), long-term vector memory (pgvector), and multi-user RBAC. AGPL-3.0

Zero-dependency Go CLI / package manager that installs Agent Skills and MCP servers into Gemini CLI (and other tools), plus Claude Code slash commands and hooks

Finds API keys in the chats, logs and checkpoints Gemini CLI keeps under ~/.gemini/tmp and in the MCP servers of ~/.gemini/settings.json, redacts them in place, and installs BeforeAgent / BeforeTool / SessionEnd hooks (spillage guard install) that block a prompt with a key in it and reads of .env. Python, no dependencies, no network; also covers Claude Code, Codex CLI and Cursor

Portable Claude Code setup snapshots — export your settings, plugins, hooks, CLAUDE.md, and MCP configs as a .tar.gz, diff before applying, and restore on another machine in under 2 minutes. No network, no daemon, .bak safety on every apply. Node.js 18+, cross-platform (macOS + Linux). Install: /plugin marketplace add adhenawer/claude-snapshot

Local policy layer with a BeforeTool hook for Gemini CLI (openmoat init) that checks every shell command, file read and write, web request and MCP call against one policy before it runs, blocking what the policy denies. Adds an egress proxy, MCP server pinning and a hash-chained audit log, and the same policy covers Claude Code, Codex, Cursor and other agents. Rust, MIT OR Apache-2.0

Safety-check MCP servers and AI agents against the OWASP agentic Top 10, and get the safest setup to run them. Install with npx skills add safetycard/safetycard --agent gemini-cli

Auto-configures Claude Code hooks, MCPs, and skills for any project. Scans 95+ project types, selects from 28+ hooks, scores quality (0-100), translates configs across AI tools. Safe: dry-run, atomic writes, backup + rollback

The first social network for AI agents. Connect any AI agent via MCP to post, chat, follow other agents, discover experts, and build reputation. 10 MCP tools, open source

Open-source public home for independent AI agents: rooms, replies/@mention inbox with webhook pushes, knowledge base, job board, agent directory and a human-help queue over MCP, A2A and REST. Self-hostable on Cloudflare Workers + D1. MIT

Public need/have list for AI agents: an agent posts what it needs or has, and others reply in public under the post. No accounts and no matcher. Remote MCP server (7 tools) plus JSON API with OpenAPI. Self-hostable on Cloudflare Workers + D1. MIT

Agent-to-agent marketplace with escrow, negotiation, and reputation for off-chain and on-chain (EVM) settlement