One-liner install command per agent, plus the bypass-proof plugin where available.

Install matrix

This page is generated from integrations/agents.json — the single source of truth for every Treeship integration. Machine-readable mirror: /.well-known/treeship-agents.json.

Tier definitions (release_status):

🟢 shipped — installable and end-to-end verified today.
🟡 candidate — works via the documented install, vendor marketplace submission pending.
⚪ unavailable — this tier does not exist because the agent's runtime does not expose the required surface.

The matrix

Agent	Skill	MCP	Plugin (hook-based)	Plugin status
Claude Code	`treeship add claude-code`	`treeship add claude-code`	`claude plugin marketplace add zerkerlabs/treeship && claude plugin install treeship@treeship`	🟢 shipped
OpenClaw	`treeship add openclaw`	`treeship add openclaw`	`openclaw plugins install @treeship/openclaw-plugin`	🟢 shipped
Codex CLI	`treeship add codex`	`treeship add codex`	`npx skills add zerkerlabs/treeship --skill treeship-dev --agent codex -g -y`	🟡 candidate
Kimi Code CLI	`npx skills add zerkerlabs/treeship --skill treeship --agent kimi-cli -g -y`	`kimi mcp add --transport stdio treeship --env TREESHIP_ACTOR=agent://kimi-code -- npx -y @treeship/mcp`	`treeship integrations install kimi-code`	🟡 candidate
Cursor	`treeship add cursor`	`treeship add cursor`	—	⚪ unavailable
Hermes	`treeship add hermes`	`see integrations/hermes/README.md (yaml config with TREESHIP_ACTOR=agent://hermes)`	—	⚪ unavailable

Skill — drops a SKILL.md instruction file into the agent's skills directory. The agent reads it and follows the instructions to call treeship CLI commands. Bypass-able by prompt injection — the same agent being attested is the one deciding whether to attest.
MCP — exposes Treeship as MCP tools the agent can call. Captures every MCP-routed tool call automatically. Stronger than skill alone (no agent discipline required for MCP-routed work) but doesn't cover the agent's built-in non-MCP tools.
Plugin (hook-based) — runs below the agent in the host process. Pre/post-tool-call hooks fire regardless of what the agent says, remembers, or has been prompt-injected to do. Bypass-proof. This is the integrity gold standard.

curl -fsSL treeship.dev/setup | sh
treeship init
treeship add       # detects and instruments every supported agent on this machine

Or pick one:

Surfaces where the vendor controls the runtime and a third-party plugin is not currently possible:

Kimi (web / kimi.com) (Moonshot AI) — no_public_surface. No public third-party plugin SDK. Skills on kimi.com are user-uploaded docs, not executable code.
Kimi (iOS / Android) (Moonshot AI) — no_public_surface
Kimi Agent Swarm (K2.6+) (Moonshot AI) — api_only. OpenAI-compatible API at platform.moonshot.ai. No hook-injection below the model's instruction surface.
Perplexity Computer / Comet (Perplexity) — partner_shipping. Perplexity shipping the official Treeship Computer skill (per founder, May 2026). Receipt agent_name: perplexity-computer.
Perplexity Pro Search (Perplexity) — no_public_surface. First-party Connectors only; no third-party connector SDK.