Integrations

AgentXchain is agent/IDE/LLM agnostic by design. The five adapters (manual, local_cli, api_proxy, mcp, remote_agent) support virtually every connection pattern. These guides show you exactly how to connect your preferred platform.

How it works

Every agent connects through a runtime configured in your agentxchain.json. Each runtime uses one of five adapter types:

Adapter	When to use	Transport
manual	Human-in-the-loop roles	Operator reads prompt, writes result file
local_cli	Local CLI tools (Claude Code, Codex, Cursor)	Subprocess stdin/argv
api_proxy	Direct LLM API calls	HTTP to provider endpoint
mcp	MCP-compatible agent servers	stdio or Streamable HTTP
remote_agent	External agent services (Devin, custom)	HTTP POST/response

Pick the guide for your platform below. Each is standalone — you only need the guides relevant to your setup.

---

IDE / Agent Platforms

• Claude Code — local_cli via claude -p
• OpenAI Codex CLI — local_cli via codex
• Cursor — local_cli
• VS Code — Extension + local_cli
• Windsurf (Codeium) — local_cli
• Google Jules — api_proxy (Google)
• Devin — remote_agent (HTTP)

Local Model Runners

• Ollama — api_proxy via localhost:11434/v1
• MLX (Apple Silicon) — api_proxy via mlx-lm.server

API Providers

• Anthropic — Claude Opus 4.6, Sonnet 4.6, Haiku 4.5
• OpenAI — GPT-5.4, GPT-5.3-Codex, GPT-OSS
• Google — Gemini 3.1 Pro, Flash, Gemma 4
• DeepSeek — V3.2, R2, Coder-V3
• Mistral AI — Devstral 2, Codestral, Leanstral
• xAI — Grok 4.20 Beta 2, Grok Code Fast 1
• Amazon Bedrock — Nova 2 Pro, Nova 2 Lite, Nova Premier
• Qwen (Alibaba) — Qwen3-Coder-480B, Qwen3.6-Plus
• Groq — Ultra-fast inference for GPT-OSS, Qwen3, Llama 4
• Cohere — Command A Reasoning, Command R+

Protocol Native

• MCP (Model Context Protocol) — Any MCP-compatible agent via stdio or HTTP