Skip to content

Integrating Your Agent (OpenClaw, Pi, MCP)

This guide is for teams that want confidence integrating an existing agent runtime with Sage.

What “Confident Integration” Means

You can answer yes to all of these:

  • Your agent can discover relevant skills/prompts before generating from scratch.
  • Your agent can run in read-only mode first (no accidental publish/governance writes).
  • Your team can choose escalation level: suggest only, draft changes, or execute approved actions.
  • You have a clear fallback if Sage tools are temporarily unavailable.

How People Are Using OpenClaw and Pi

Common patterns from OpenClaw and Pi ecosystems:

  • OpenClaw users install plugins, use hooks, and optionally route external runtimes through ACP.
  • Pi users run either CLI extension mode (fastest path) or integrate at harness/SDK level for deeper control.
  • Teams often start with low-risk read/search integration first, then add proposal/publish workflows.

Path A: Runtime-Agnostic MCP Integration (Fastest)

If your runtime supports MCP, start here:

sage mcp start
sage mcp list-tools

Use your runtime MCP config to point to:

  • command: sage
  • args: ["mcp", "start"]

This gives agent-side discovery/search without requiring plugin coupling.

Path B: OpenClaw Integration

Install Sage plugin:

openclaw plugins install @sage-protocol/openclaw-sage
openclaw plugins info openclaw-sage

What this typically enables:

  • Sage tool bridge inside OpenClaw sessions
  • Session hooks for context/suggestion workflows
  • Skill-aware discovery and execution patterns

Rollout recommendation:

  1. Enable discovery-only behavior first.
  2. Add proposal drafting after your team validates outputs.
  3. Keep execution gated by governance policy.

Path C: Pi Agent Integration

CLI extension mode:

pi install npm:@sage-protocol/pi-adapter
pi list

Harness/SDK mode is useful when you own the Pi session bootstrap and need strict control of tool registration and lifecycle.

Integration Confidence Checklist

Before production use, validate:

  1. Discovery works
  2. Agent can find relevant skills/prompts and cite source context.
  3. Safety gates exist
  4. Write actions are disabled or human-approved in early rollout.
  5. Failure mode is clear
  6. If Sage is unavailable, agent falls back gracefully and reports degraded mode.
  7. Governance boundary is explicit
  8. Agent can suggest and draft, but execution follows your DAO/operator policy.
  • Start with discovery: agent reads, searches, and cites governed sources.
  • Add draft authority: agent proposes updates, packages changes, or drafts governance actions.
  • Add governed execution: agent operates under explicit governance and runbook guardrails.

For the broader loop, see Protocol Deep Dive.