Agent Integration

AVC integrates with AI agent frameworks via the Model Context Protocol — every CLI command is exposed as a tool that agents can call directly.

One-shot setup

avc init --skills claude-code,cursor,windsurf

This writes the right config and instruction files for each framework into your project. No further setup needed — open your editor, the agent finds AVC automatically.

Pages in this section

MCP Server How AVC exposes its 14 commands as MCP tools. Tool list, manual config, troubleshooting.

Claude Code .claude/settings.json + skill files. Workflow walkthrough and troubleshooting.

Cursor .cursor/mcp.json + rules file. Workflow walkthrough and verification.

Windsurf .codeium/windsurf/mcp_config.json + .windsurfrules section.

How agents use AVC

A typical agent workflow with AVC:

1. Snapshot the current state before any non-trivial change (avc_snapshot)
2. Branch if the work is uncertain or experimental (avc_branch_create)
3. Make changes inside the workspace
4. Snapshot the result
5. Diff against the baseline to summarize what changed (avc_diff)
6. Restore or merge based on user approval

The skill / rules files written by avc init --skills instruct the agent on when to use each tool.

Custom integrations

If your agent framework isn’t covered by --skills, use the generic preset:

avc init --skills generic

This writes AGENT_INSTRUCTIONS.md — a framework-neutral document you can include in any agent prompt.

For MCP-aware frameworks, the minimal config is:

{
  "mcpServers": {
    "avc": {
      "command": "avc",
      "args": ["mcp", "serve", "--compact"]
    }
  }
}

See MCP Server for the full tool catalog and protocol details.