One skill to rule them all.

GearCore is a unified skill & MCP hub for AI CLIs. Register your MCP servers and skills once, expose them as a single native skill in Claude Code, Codex, Kimi, and OpenCode — and keep every context window lean.

Get started View source ↗

The problem

MCP config sprawl doesn't scale

Every AI tool wants its own copy of the same server configs. Every project repeats them. Contexts bloat, configs drift, and nothing stays in sync.

# Without GearCore
Claude:   12 servers × schema tokens, copied into every project
Codex:    same configs, copied again
Kimi:     same configs, copied again

# With GearCore
Claude → gearcore → request_skill("web-research")
Kimi   → gearcore → request_skill("filesystem")
Codex  → gearcore → request_skill("memory")

🎭 Native skill

AI tools discover gearcore through their own skill mechanism. No per-tool config duplication, no plugins to install.

🔒 Progressive disclosure

Backend tools stay hidden until you explicitly unlock a bundle with request_skill. Context stays clean by default.

📁 Project scoping

Each repo can allowlist exactly the skills it needs via .gearcore/config.yaml. Global power, local discipline.

⓪ Level-0 skills

Mark core skills to be revealed by default — full instructions inline in list-skills, embedded at sync time.

⚔️ Conflict resolution

Same tool name from multiple backends? GearCore deduplicates, namespaces, or unifies automatically.

🔄 One-command updates

gearcore update refreshes MCP servers, skill bundles, vendored skills, and re-syncs everything — version-aware, with dry-run.

Install

Up and running in two commands

Requires Python 3.13+ and uv.

# 1. Install the CLI from a clone
git clone https://github.com/BlinkVoid/GearCore.git && cd GearCore
uv tool install .

# 2. Sync the self-skill to every detected AI CLI
gearcore sync

# Register an MCP server once…
gearcore add-mcp --id filesystem --type stdio \
  --command npx --args -y @modelcontextprotocol/server-filesystem ~/workspace

# …and check what your tools will see
gearcore status
gearcore list-skills

Compatibility

Verified across four AI CLIs

Each client gets the self-skill installed at its native discovery path, verified end-to-end with live hub handshakes.

Client	Skill install target	Discovery	Hub handshake
Claude Code	`~/.claude/skills/gearcore`	✓	✓
Codex CLI	`~/.codex/skills/gearcore`	✓	✓
Kimi CLI	`~/.kimi/skills/gearcore`	✓	✓
OpenCode	`~/.config/opencode/skills/gearcore`	✓	✓