Loading ToolBench
MCP server for the Prompt Compiler application, providing tools to optimize, compile, and generate prompts and agent packs using the myCompiler API backend
Server has 8 tools with basic schemas and descriptions, but significant quality gaps. All tools have descriptions (10-200 chars, mostly adequate), and input schemas are present with type definitions. However, descriptions lack actionable context for LLM selection (no WHEN to use, no prerequisites, no recovery guidance). Parameter descriptions are minimal or absent. Output schemas are completely undocumented, LLMs cannot plan downstream calls. Error handling is not visible in the code. Two tools (plan_agent_pack, apply_agent_pack) perform file system operations but lack confirmation/dry-run patterns. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Naming is reasonable (verb_noun pattern) but some tools are vague: 'optimize_prompt' vs 'compile_prompt' distinction is unclear.
Generate and WRITE a repo-aware Claude agent pack into the local repository. Existing files are never silently overwritten: a conflicting path is written as ``<path>.new`` unless its path is included in ``overwrite`` (which the plan surfaces).
Benchmark a raw prompt against the compiled prompt and return the comparison payload.
Compile a prompt through the Prompt Compiler backend and return the structured response.
Convert a generated agent prompt into a Claude Code project pack manifest.
Generate a Claude-friendly agent/system prompt from a natural-language description.
Generate a structured skill/tool definition from a natural-language capability description.
Output schemas completely undocumented. No tool returns a documented schema, LLMs cannot infer what fields to expect or plan chained calls. E.g., compile_prompt, generate_agent, benchmark_prompt all lack return type documentation.
Descriptions lack actionable context. No tool explains WHEN to use it vs similar tools, what prerequisites exist, or what to do on failure. E.g., 'optimize_prompt' vs 'compile_prompt', when should an LLM pick one over the other?
apply_agent_pack performs destructive file writes but lacks confirmation/dry-run pattern. No tool annotation (destructiveHint) despite WRITE risk. Agents could overwrite user files without explicit approval.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Optimize a prompt using the myCompiler API.
Preview a repo-aware Claude agent pack for the local repository (writes nothing). Reads the repo at ``path``, generates a tailored pack on the backend, and returns the file plan with create/overwrite/identical actions against existing files, plus detected stack and commands. Nothing is written to disk.
Parameter descriptions are minimal or missing. 'model' param in benchmark_prompt has a description, but most other params lack context. E.g., 'pack_type' in plan_agent_pack, what are valid values? What does each type generate?
No error handling guidance visible. No tool documents what errors can occur, whether they are retryable, or what the LLM should do next. E.g., if compile_prompt fails due to invalid syntax, should the agent retry, ask the user, or try a different tool?