Investigate performance improvements for command execution

[pento]

Summary

Commands triggered from the WordPress editor (proofread, review, etc.) can feel slow because each MCP tool call takes several seconds, and a single command may require multiple sequential tool calls. Investigate ways to reduce the end-to-end latency.

Current bottlenecks

Each command typically involves multiple round-trips:

1. wp_update_command_status (running)
2. wp_open_post (if not already open)
3. wp_read_post
4. One or more editing tool calls (wp_edit_block_text, wp_add_note, etc.)
5. wp_update_command_status (completed)

Each tool call has overhead from the MCP stdio transport, LLM inference time, and HTTP round-trips to WordPress.

Areas to investigate

• Batching: Could multiple operations be combined into fewer tool calls?
• Pre-loading: Since the command includes post_id, could post content be pre-fetched and included in the channel notification?
• Prompt design: Can prompts be structured to encourage fewer, more targeted tool calls?
• Parallel tool calls: Are there sequential calls that could be made parallel?
• Caching: Could recently-read post content be reused if the post hasn't changed?