Happy slice of mcp pizzaMCP.pizza

opencode-x

MCP.Pizza Chef: sdeonvacation

OpenCode X is a free, open-source coding agent for the terminal, built on opencode and shaped to match Claude Code feature for feature. Hooks, plugins, agents, skills and your existing CLAUDE.md carry over with no migration at all. On top of that it adds parallel helper agents, long-running goals, memory that survives between sessions, and automatic squeezing of tool output that cuts roughly a third off what you spend. No subscription — you bring your own provider keys.

Coding

Use This MCP client To

Run a coding assistant without a monthly subscription Switch between Claude, GPT and Gemini mid-task Send a long refactor to run in the background Split a big job across several helper agents Reuse my existing Claude Code hooks and skills Keep working after a session runs out of room

README

OpenCode X

npm version License: MIT GitHub stars TypeScript Built with Effect

Everything Claude Code does and more. Any model. Zero subscription. Smarter with your tokens.

OpenCode X is a terminal AI coding agent forked from opencode with 280+ commits. It matches Claude Code's headline features (autonomous goals, background execution, parallel subagents, hooks) while adding what Claude Code can't offer: provider freedom, intelligent token compression, 3-tier context safety, configurable orchestration guardrails, and an open-source MIT codebase you actually own.

Your existing claude code hooks, plugins, agents and skills work immediately. Zero migration.

OpenCode X — terminal AI agent with swarm execution and multi-provider support
Autonomous multi-turn execution with real-time cost tracking

Why Switch?

Coming from Claude Code?

You're paying $200+/month per developer for a single-provider tool. OpenCode X gives you the same workflow with genuine advantages:

  • $0/month — bring your own API keys, pay only for tokens you use
  • Any model — Claude, GPT, Gemini, Mistral, Llama, DeepSeek, or any of 75+ providers
  • 30-60% token savings — LLM compression on tool outputs before they reach your model (Claude Code sends raw output)
  • Never lose context — 3-tier safety net auto-recovers at 75%, 97%, and 413 overflow (Claude Code only has auto-compact)
  • Your hooks already work — same config, same env vars, same lifecycle events
  • Configurable guardrails — doom loop detection, spawn limits, per-model concurrency (Claude Code doesn't expose these)
  • Hybrid routing — automatically route compression/titles to cheap models, reasoning to premium ones
  • Open source — MIT, no telemetry, no account, fork and extend freely

opencode-x FAQ

Does it cost anything?
The program is free and MIT licensed; you only pay your chosen model provider for what you use.
Do my Claude Code settings still work?
Yes — hooks, plugins, agents, skills and CLAUDE.md all work with no changes.
Which models can I use?
Claude, GPT, Gemini, Mistral, Llama, DeepSeek and around 75 other providers, switchable mid-session.
Can I use this instead of a paid Claude Code subscription?
Yes — it matches the headline features and bills through your own provider keys instead.
Does it work with MCP servers?
Yes, and you can list only the tools you want from each server to keep sessions lean.
How do I install it?
It is a terminal program installed from npm, so you need Node and some comfort with a command line.
Does my code go anywhere?
There is no usage tracking and no account — your code only reaches the model provider you choose.