Loading ToolBench
MCP server exposing Alchemy infrastructure operations as tools for Claude, Cursor, and other MCP clients
8 tools with complete schemas and descriptions, but significant quality gaps. All tools have descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization) and input schemas with enums for stage/deployment parameters. However, descriptions are generic and lack actionable context (e.g., 'Get the current status' vs 'Retrieve deployment health, resource counts, and recent errors for the specified stage'). No output schemas documented. Error handling guidance absent. Destructive tools (destroy_infrastructure) lack confirmation patterns. Parameter descriptions are minimal. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels.
Deploy Alchemy infrastructure to a specific stage (prod, staging, or preview)
Destroy all Alchemy resources for a specific stage
Get statistics about KV namespace usage
Get the current status of all deployed Alchemy resources
List objects in the R2 storage bucket
Execute a SQL query against the D1 database
Send a message to a Durable Object instance
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., resource IDs for chaining). Baseline: 100% of A+ tools document return types.
Descriptions are 25-40 characters, well below the 50-200 char baseline for LLM optimization. Generic phrasing ('Get the current status', 'Execute a SQL query') lacks context for tool selection. No WHEN to use, no prerequisites, no recovery guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Start a Cloudflare Workflow execution
Destructive tool (destroy_infrastructure) lacks confirmation pattern. No dry-run option, no explicit confirmation requirement in schema. Agents can accidentally destroy production resources.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools have explicit Risk levels (READ_ONLY, WRITE, DESTRUCTIVE) in metadata but not exposed via MCP tool annotations. LLMs cannot infer safety properties.
Parameter descriptions minimal or absent. 'stage' parameter appears in 7 tools but descriptions are identical and lack context (e.g., 'The deployment stage' vs 'Deployment stage: prod=production, staging=pre-release, dev=local testing'). No guidance on which stage to use when.