AI Console

The AI Console is a developer-facing diagnostic view that shows what tools the AI is calling, when, and what they returned. It is the source of truth for understanding the AI's actions when a chat prompt produces a side effect (a new request, a new workflow, a database query, a navigation, etc.).

When to use it

• A chat reply mentions "I created a workflow" but you can't find it
• A tool call returned an error and you want the full payload
• You're debugging an MCP-tool description and want to see the AI's exact input
• You want to know whether the AI used Rostyman__send_request or fell through to a different tool

Opening the console

Open the AI panel (Sparkles icon or Ctrl+J), then switch to the MCP Tool Calls tab. The view lists every tool invocation from the current conversation, newest first.

What each row shows

Column	Meaning
Tool name	The MCP tool the AI selected (e.g. create_request, query_db)
Duration	Wall-clock latency of the tool handler
Caller	Provider + model that issued the call (Anthropic / OpenAI / Gemini / Copilot / local-cli)
Status	Green = success · Red = error

Click a row to expand its input and output (or error) JSON blocks.

Filtering

• Per-tool filter — dropdown in the header restricts the list to a single tool. Useful when chasing a bug in one tool's schema.
• Load 200 more — pagination button at the bottom; press to fetch the next page when more entries exist.

Truncation

Tool input and output are truncated to the value of the audit.maxBodyLength setting (default 4000 chars). Truncated payloads end with ...(truncated) so it's clear when content was cut. Raise the limit in Settings → Advanced if you need full bodies for compliance reasons.

Retention

Tool calls live in the mcp_tool_call_log SQLite table. A boot-time sweep deletes rows older than audit.retentionDays (default 90 days).

Related

• AI History — every AI call (prompt → response), not just tool calls
• MCP Server — the registry of tools the AI can invoke
• MCP Overview — protocol-level explanation