rps321321/obsidian-mcp-pro

Better agent UX on every tool

No runtime behavior changed in this release — but every MCP client and AI agent interacting with obsidian-mcp-pro will now see substantially better tool metadata. The improvements help agents pick the right tool, pass valid arguments on the first try, and interpret results correctly.

What changed

• All 23 tool descriptions rewritten to convey what's returned, when to use each tool, and how they compose. Previously many were one-liners like "Find all notes that link to a specific note"; they now describe output shape, use cases, edge cases, and cross-references.
• Zod schemas tightened with .int(), .min(), .max(), and regex constraints where applicable. Every parameter now has a .describe() with concrete examples and default values.
• title field added to every tool — a human-readable display name separate from the machine-readable tool name.
• annotations added to every tool (readOnlyHint, destructiveHint, idempotentHint, openWorldHint). Well-behaved clients can now surface confirmation UIs before destructive operations like delete_note and move_note, and safely cache read-only tool results.

Measured impact

On Glama's tool quality scorer:

| Metric | Before | After |
|---|---|---|
| Average score | 3.14 / 5 | 4.40 / 5 |
| A-grade tools | 2 / 23 | 23 / 23 |
| Biggest single lift | — | list_canvases 3.1 → 4.7 |

For contributors

A new internal guide — docs/TOOL_AUTHORING.md — documents the description, schema, and annotation conventions so new tools start at A-grade.

Install

npx obsidian-mcp-pro        # npm install -g obsidian-mcp-pro

Full changelog: https://github.com/rps321321/obsidian-mcp-pro/blob/master/CHANGELOG.md#112---2026-04-15