Loading ToolBench
CADI MCP (Model Context Protocol) server for LLM integration with content-addressed chunk storage, retrieval, and code analysis
CADI MCP server has well-structured tool definitions with complete JSON schemas and descriptions for all 10 tools. Naming follows verb_noun convention (cadi_search, cadi_get_chunk, cadi_import). However, descriptions are inconsistent in depth and actionability. Most descriptions lack explicit guidance on when to use each tool vs. alternatives, and several omit error recovery hints. Parameter descriptions are present but often generic. Output schemas are not documented in the provided source. Error handling is mentioned as a feature but no specific error recovery patterns are visible in tool definitions.
Build a CADI manifest for a specific target. Assembles chunks into runnable project.
Explain a chunk's purpose, dependencies, and lineage. Useful for understanding what code does.
Retrieve chunk content by ID (~100 tokens). Use after search/resolve to get the actual code.
⚡ IMPORT PROJECT ONCE, REUSE FOREVER. Analyzes codebase, creates chunks with aliases. Do this first with any new project!
Show the build plan for a manifest without executing it
Publish chunks to registry for team sharing. Share solutions so others don't rewrite them.
⚡ FAST LOOKUP (~30 tokens). Resolve a human-readable alias to chunk ID. Use for known chunks like 'myproject/utils/logger'.
Output schemas not documented. Tool definitions show input schemas but no return type documentation. LLMs cannot plan downstream calls or extract required fields (e.g., does cadi_search return chunk_id, content, metadata?).
Descriptions lack actionable guidance on tool selection. 'Get AI-powered suggestions' (cadi_suggest) does not explain when to call it vs. cadi_search. No hints like 'Use this when you don't know what to search for' or 'Call search first for known keywords.'
cadi_publish has no required parameters (empty required array). This is unusual for a write operation and may indicate incomplete schema definition or overly permissive defaults that could cause unintended publishes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
⚡ SEARCH FIRST! Find existing code chunks (~50 tokens vs 500+ to write new). Search by keyword, concept, or language.
Get AI-powered suggestions for chunks that might help with a task. Good when you're not sure what to search for.
Verify a chunk's integrity and provenance
Error handling patterns not visible in tool definitions. No guidance on retryable vs. fatal errors, or recovery steps (e.g., 'If chunk not found, try cadi_search first').
cadi_plan and cadi_build descriptions are vague about differences. Both take manifest and target; unclear when to call plan vs. build without reading implementation.