Skip to content

Using the MCP Server

Use this guide when your primary goal is agent integration confidence, not CLI-first workflows.

Why This Matters

MCP gives your agent direct access to Sage discovery and workflow tools through a stable protocol boundary.

Where This Fits in the Loop

This guide is part of Discover and Use.

Step 1: Start Sage MCP

sage mcp start

For MCP runtimes, configure Sage as:

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

Step 2: Verify Tool Availability

sage mcp list-tools

You should see tool families like discovery, skills, prompts, governance, and hub operations.

Step 3: Smoke Test a Tool Call

sage mcp call get_project_context --args '{}'

If this returns JSON successfully, your MCP bridge is working end-to-end.

Step 4: Use Read-Only First

Start with read-only discovery tools before enabling write paths:

  • search_skills
  • search_prompts
  • list_libraries
  • get_prompt

This reduces integration risk while you validate relevance and output quality.

Step 5: Add Write Paths Deliberately

After read-only confidence:

  • Enable draft/propose workflows.
  • Keep execution behind human or governance approval.
  • Add clear policy for when agents can publish, propose, or execute.

Integration Modes

  • Assisted: MCP read/search only; humans decide all writes.
  • Collaborative: agents draft/suggest; humans execute.
  • Autonomous: agents act within explicit governance/runbook guardrails.

For full runtime-specific guidance: - Integrating Your Agent

Troubleshooting

If tools do not appear or calls fail:

sage mcp list-tools
sage mcp call hub_status --args '{}'
sage doctor