MCP

Use MCP when you want Claude Code, Cursor, or another MCP client to try Aionis without rewriting the host Agent loop first.


npx @aionis/mcp@latest --base-url http://127.0.0.1:3001 --scope my-project

The MCP bridge talks to a running Aionis Runtime and exposes a small set of tools for memory, execution context, governance, measurement, and replay.

Claude Code Example


claude mcp add --transport stdio --scope project aionis -- \
  npx -y @aionis/mcp@latest \
  --base-url http://127.0.0.1:3001 \
  --scope my-project

Use a project-specific scope so memory from one repository does not leak into another.

For a shareable walkthrough, use the Claude Code Demo.

Available Tools

Tool	Purpose
`aionis_context`	Compile governed execution memory for the current Agent turn.
`aionis_record_step`	Record a planner/worker/verifier/reviewer step and optional feedback.
`aionis_handoff`	Record branch-aware handoff for cross-session continuation.
`aionis_remember`	Store ordinary project memory.
`aionis_govern_memory`	Govern external memory candidates through admission surfaces.
`aionis_measure`	Measure guide and feedback impact for a run.
`aionis_snapshot`	Return an operator snapshot.
`aionis_flight_recorder`	Replay what memory the Agent could see at decision time.
`aionis_health`	Check Runtime reachability.

Typical Coding-Agent Flow

Call aionis_context before a task turn to get governed context.
Run the Agent with the returned context.
Call aionis_record_step after meaningful actions or verifier results.
Call aionis_handoff at session boundaries.
Call aionis_measure and aionis_snapshot for operator review.

`aionis_context`

Use this when the Agent asks “what should I know before continuing?”


{
  "run_id": "run-001",
  "task_signature": "checkout-migration",
  "agent_id": "worker-1",
  "role": "worker",
  "query_text": "Continue the checkout migration from the accepted path.",
  "context_mode": "compact_agent",
  "repo_state": {
    "existing_files": ["src/checkout/adapter.ts"],
    "missing_files": ["src/checkout/bundledDev.ts"]
  }
}

The repo_state field is optional but useful. It lets Aionis render missing active targets as pending work instead of letting the Agent silently fall back to old files.

`aionis_record_step`

Use this after tool calls, test runs, verifier results, or branch decisions.


{
  "run_id": "run-001",
  "task_signature": "checkout-migration",
  "agent_id": "verifier-1",
  "role": "verifier",
  "title": "Verifier rejected legacy route",
  "summary": "The legacy adapter touched unrelated modules.",
  "outcome": "failed",
  "target_files": ["src/checkout/legacyAdapter.ts"]
}

MCP Boundary

MCP is the easiest trial path, but it is still a bridge. Your MCP client owns the Agent loop and tool execution. Aionis owns memory governance and audit surfaces.

Guide: AIONIS_MCP.md .