You are a draw.io diagram editor. Use these tools to read, modify, and create draw.io (.drawio / .xml) diagram files.
Server has 11 tools with consistent naming (verb_noun pattern) and basic descriptions. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are generic and lack LLM-optimization guidance (e.g., 'when to use', 'what happens next'). Output schemas are undocumented, LLMs cannot predict return structure. Error handling is minimal; no recovery guidance or categorization. Missing tool annotations (readOnlyHint, destructiveHint). Parameter constraints lack enums where appropriate (e.g., 'role' in add_device, 'direction' in auto_layout, 'style_profile'). No pagination support despite tools returning lists. Baseline: avg tool description ~100 chars (below 194 baseline); all params have descriptions but lack constraint details.
Add a network device node to a draw.io diagram. The device carries structured metadata (hostname, role, vendor, platform, layer, site, zone, redundancy_group) stored in the draw.io tooltip field so it survives file round-trips.
Add a directed edge (arrow/connector) between two nodes.
Add a new shape (vertex/node) to a draw.io diagram.
Auto-layout the diagram using a layered graph approach.
Create a new blank draw.io diagram file.
Delete a node or edge from the diagram by its cell ID. Also removes any edges connected to the deleted node.
List all draw.io diagram files (.drawio, .xml) in a folder.
Output schemas undocumented. Tools return strings (JSON or plain text) but LLMs cannot predict structure. E.g., list_diagrams returns JSON array, list_nodes returns JSON with cell objects, but no schema is declared. LLMs cannot plan downstream calls or extract fields reliably.
Missing tool annotations. No readOnlyHint, destructiveHint, or idempotentHint declared. Agents cannot distinguish safe (read_diagram) from destructive (delete_node) operations without parsing descriptions. Violates current MCP spec (2026-07-28).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
List all nodes (shapes/vertices) and edges in a draw.io diagram.
Read the raw XML content of a draw.io diagram file.
Update the label, style, or geometry of an existing node. Only the fields you provide will be changed.
Write (overwrite) a draw.io diagram file with new XML content. Creates the file if it does not exist.
Enum constraints missing. 'direction' in auto_layout accepts TB|BT|LR|RL but is declared as free-form string. 'role' in add_device has 18 valid values but no enum constraint. 'style_profile' in add_device (minimal|enterprise|dark|vendor-neutral) lacks enum. LLMs will hallucinate invalid values.
Error handling lacks recovery guidance. Tools return plain error strings (e.g., 'ERROR: Not a directory') with no actionable next steps. No categorization of retryable vs user-fixable errors. Agents cannot self-correct or know whether to retry.
Descriptions lack LLM-optimization. Generic docstrings (e.g., 'Add a new shape (vertex/node) to a draw.io diagram') do not explain WHEN to use the tool vs alternatives, WHAT happens on success/failure, or WHAT the agent should do next. Baseline: descriptions should be 50-200 chars with clear intent signals.