mcp-drill scan results: 27 tools, 0 output schemas, 0% enforceable contracts

## mcp-drill scan — mcp-server-chart 0.8.x

| Metric | Result |
|--------|--------|
| Protocol | 2025-06-18 |
| Tools | 27 |
| Output-schema coverage | 0% (0/27) |
| Enforceable contracts | 0% (0/27) |
| Error handling | 100% (3/3 probes clean) |

### What this means

mcp-server-chart has the best error handling we've seen — every bad-input probe returns a clean JSON-RPC error. 27 tools, no crashes, no hangs.

The gap is output schemas: none of the 27 tools declare an `outputSchema`, so downstream agents have no way to validate whether a chart response is structurally correct before rendering or acting on it. For a visualization server this matters — an agent passing a malformed chart spec to a renderer could produce silent failures.

### Why this matters for chart generation

Chart tools return structured data (SVG, config objects, data specs). Adding `outputSchema` with specific property constraints would let agents validate that a chart response contains the expected format, dimensions, and data bindings — before rendering.

### How we tested

[mcp-drill](https://github.com/TimurRakhmatullin86/mcp-drill) is an open-source MCP testing harness (no LLM needed). It probes fault handling and schema hygiene across the MCP protocol.

```
pip install mcp-drill
mcp-drill scan -- npx -y @antv/mcp-server-chart@latest
```

Full methodology: Apache-2.0, 14 tests, deterministic and reproducible.

Happy to help draft schemas for the most commonly used tools if useful.

Start by inspecting the 27 chart tool definitions and how their responses are currently exposed, then run the documented mcp-drill scan command against @antv/mcp-server-chart. Add outputSchema declarations with constraints appropriate to each chart response, and rerun the scan to confirm schema coverage and enforceable contracts are no longer zero.

Written by the indexing model from the issue text.