Loading ToolBench
Read-only MCP server exposing dashboard APIs for n8n workflow inspection, error tracking, instance management, secrets metadata, MCP server configuration, and notes vault access
31 tools with mixed quality. Strengths: all tools have descriptions (10 - 200 chars, mostly good); most parameters have types and descriptions; clear verb-noun naming (list_, get_, trigger_, etc.). Weaknesses: significant duplication (list_workflows, get_workflow, list_executions appear 3+ times each); no output schemas documented; error handling is minimal; no tool annotations (readOnlyHint, destructiveHint); workspace tools lack dependency hints; no pagination guidance despite list tools returning up to 50 items. Naming is generally strong (verb-first, action-clear), but duplication suggests incomplete tool consolidation. Parameter descriptions are adequate but lack format constraints (e.g., workflow_id format, status enum values). Security: no secrets in params (good), but no audit trail or permission-gate patterns visible.
List recent errors for ONE workflow id. Use to judge whether a failure is a one-off or a recurring pattern.
Get the HA summary for the active instance: health_status, workflow_count, error_count_24h, last_execution_at. Use for context on whether this error is part of a broader outage.
Fetch a single execution by id, including its run data / error detail. This is where the actual failure payload and stack live; inspect it before proposing a fix.
Get details of a specific execution including error info and node results. Use when diagnosing a specific failed execution.
Get recent workflow errors from the dashboard error log. Use when diagnosing problems.
Dashboard's own status — configured, active_instance, version.
Severe tool duplication: list_workflows, get_workflow, list_executions, get_execution, and get_recent_errors each appear 2 - 3 times with identical or near-identical signatures. This forces LLMs to reason about which variant to call, wastes tokens, and increases error likelihood. Consolidate into single canonical tools.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., workflow_id from list_workflows to pass to get_workflow). Add explicit return type documentation for each tool.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe reads from destructive writes without explicit hints. Mark trigger_workflow, set_workflow_active, import_workflow, workspace_write, workspace_append, workspace_archive as destructive; mark all list_* and get_* as readOnly.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Get details of a specific workflow including nodes, webhook URL, and tags. Use when the user asks about a specific workflow.
Fetch full workflow definition (nodes + connections) by id.
Fetch a workflow's definition (its nodes, connections, settings) from the active n8n instance. Use to understand the node that failed and what feeds it.
Create and import a new workflow into the active n8n instance. Use when the user asks you to build, create, or set up a new workflow. Provide the full workflow JSON with nodes and connections.
Recent webhook-reported errors, newest first. `hours` filters to the last N hours; 0 disables the time filter.
List recent workflow executions with status and timing. Use when the user asks about recent runs or execution history.
List recent n8n executions. Optional filters: workflow_id (exact) and status (one of: success, error, running, canceled, waiting).
List recent executions, optionally filtered by workflow_id and status (e.g. 'error'). Use to see how often a workflow runs and how often it fails.
Configured MCP servers on this AgeniusDesk instance.
Recent webhook-posted dashboard messages (toasts). Same data the `messages` WebSocket event carries.
All configured n8n instances. Omits the encrypted api_key but includes url, login_url, and a key_hint for identification.
List the most recent n8n workflow errors AgeniusDesk has collected. Returns rows with id, workflow_id, workflow_name, execution_id, node_name, error_message, error_type, occurred_at. Start here to pick what to triage.
Names and types of stored secrets. Never returns actual values; use the Secrets UI for that.
List all n8n workflows with their names, active status, and trigger type. Use this when the user asks about their workflows.
List n8n workflows on the active instance. Returns id, name, active, tags, updatedAt for each. Use active_only=True to see running ones only or name_contains to search by substring.
Read a single note's raw markdown by its vault-relative path (e.g. 'user/runbook.md'). Returns {path, content} or an error message.
Search the notes vault by full text and/or tag. Empty query returns recent notes ordered by mtime. Tag match is case-insensitive and exact. Results include a highlighted snippet around the matched term.
Activate or deactivate an n8n workflow. Use when the user asks to enable, disable, turn on, or turn off a workflow.
Trigger/run an n8n workflow. Use when the user asks to run, trigger, or execute a workflow.
Append content to a workspace file (creates it if missing). Good for accumulating scratchpads or logs.
Soft-delete a workspace file by moving it to the archive. Never hard-deletes; the file can be recovered.
List files and folders in the workspace (the harness filesystem). Paths are relative to the workspace root. Conventions: user/ human notes, agent/ your scratchpads, docs/ documentation, workflows/ saved n8n workflow JSON, research/ add-in output, shared/ canonical facts, sessions/ session logs. AGENTS.md at the root holds the instructions that steer all agents.
Read a file from the workspace and return its full contents.
Full-text search across the workspace (title, body, tags). Returns matching paths with snippets.
Create or overwrite a file in the workspace. Creates parent folders. Use this to save workflows, documentation, or notes. Paths are relative to the workspace root. Conventions: user/ human notes, agent/ your scratchpads, docs/ documentation, workflows/ saved n8n workflow JSON, research/ add-in output, shared/ canonical facts, sessions/ session logs. AGENTS.md at the root holds the instructions that steer all agents.
Pagination guidance missing. list_* tools accept limit (default 10 - 50) but no documentation of total count, next_cursor, or pagination strategy. Large result sets risk context window exhaustion. Add pagination metadata to responses.
Error handling lacks recovery guidance. No tool documents what to do on failure (retry, ask user, abort). E.g., trigger_workflow should explain: 'If workflow not found, call list_workflows() to verify the ID.' Add actionable error messages.