Agent Compatibility

Waypoint has two separate concerns per agent: capture (automatic session recording via hook) and retrieval (querying context via MCP). Retrieval works with any MCP-compatible agent. Capture depends on whether the agent exposes a lifecycle hook and whether Waypoint can parse the transcript format it produces.

Compatibility overview

Claude Code

Surfaces: CLI, desktop app, VS Code extension, JetBrains extension

Claude Code is available across multiple surfaces. All of them share the same .claude/settings.json hook mechanism, so waypoint init covers every surface automatically — no different setup depending on whether you use Claude Code in the terminal or inside an IDE.

Capture: The Stop hook fires after every agent turn. The hook payload includes transcript_path, which Waypoint reads and parses directly.

Retrieval: get_waypoint_context is available immediately after waypoint init by connecting to your Waypoint server's MCP endpoint.

Tool mapping:

VS Code GitHub Copilot

Surfaces: VS Code (IDE)

VS Code introduced agent hooks in April 2026, currently in Preview. The hook system reads .claude/settings.json by default — the same file Waypoint writes during waypoint init — so no extra configuration is needed.

Retrieval: MCP support is available in VS Code by configuring the Waypoint server MCP endpoint.

Tool mapping:

Cursor

Surfaces: IDE

Cursor has had a full hook system since v1.7 (October 2025), configured via $REPO/.cursor/hooks.json or ~/.cursor/hooks.json. A stop hook fires when the agent finishes a task, and the payload includes transcript_path.

The remaining question is transcript format: Cursor writes JSONL with its own step structure. Whether Waypoint's parser handles it without modification is unverified. Until that's confirmed, treat Cursor capture as likely-but-not-yet-tested.

waypoint init does not currently write a Cursor hook config. To wire it up manually, add the following to $REPO/.cursor/hooks.json:

{
  "stop": [
    { "type": "command", "command": "waypoint generate" }
  ]
}

Retrieval config (~/.cursor/mcp.json):

{
  "mcpServers": {
    "waypoint": {
      "type": "http",
      "url": "https://your-waypoint-server.com/api/mcp",
      "headers": {
        "Authorization": "Bearer wp_..."
      }
    }
  }
}

Windsurf

Surfaces: IDE

Windsurf's Cascade exposes a post_cascade_response_with_transcript hook, but it fires after every response turn — there is no session-end equivalent. Each invocation writes the full conversation transcript (growing across turns) to ~/.windsurf/transcripts/{trajectory_id}.jsonl. Integrating this with Waypoint would require reading the growing file and deduplicating events across invocations, which is a different model than the stop hook approach.

Retrieval config (~/.codeium/windsurf/mcp_config.json):

{
  "mcpServers": {
    "waypoint": {
      "type": "http",
      "url": "https://your-waypoint-server.com/api/mcp",
      "headers": {
        "Authorization": "Bearer wp_..."
      }
    }
  }
}

Other MCP-compatible agents

Any agent that supports MCP tools can call get_waypoint_context for retrieval. For capture, waypoint generate accepts a transcript via stdin — if you can export a session transcript from your agent, you can pipe it in manually or wire it to whatever lifecycle hook your agent exposes.

Agent-agnostic design

Session records store the agentModel field but are otherwise model-neutral. A session captured during a Claude Code session is fully readable during a Cursor or Windsurf session querying via MCP. The context layer is portable with the repository regardless of which agent produced it.