Loading ToolBench
MCP server for indexing codebases using Ollama embeddings and Qdrant vector database
This MCP server has two tools with significant quality gaps. Both tools have basic descriptions (30-40 chars) and minimal parameter documentation. The schema for indexCodebase includes only a 'dir' parameter with a one-line description ('Directory to index'); searchCodebase has slightly better documentation with 'query' (required) and 'topK' (optional with default=5) parameters. However, neither tool's description clearly states WHAT happens when called (output format, side effects), WHEN to use it vs alternatives, or what the LLM should expect in response. Parameter descriptions lack constraints, ranges, or format guidance. No output schemas are documented anywhere in the source code. Error handling is declared as 'errorReporting: true' but no actual error recovery guidance is visible in the tool definitions. The server reads the codebase from disk (READ_ONLY risk), but descriptions don't explain indexing cost, latency expectations, or limitations. Overall, the tools are functional but poorly described for LLM consumption.
Index all files in a directory
Semantic search over codebase
Tool descriptions are too brief (30-40 chars) and lack essential context. 'Index all files in a directory' and 'Semantic search over codebase' do not explain output format, when to call them, what the LLM should expect, or how they differ from other potential indexing/search approaches.
Output schemas are completely undocumented. Callers do not know what structure indexCodebase or searchCodebase return. Without documented return types, LLMs cannot plan downstream operations or extract the right fields.
Parameter descriptions are minimal or missing context. 'dir' is described as 'Directory to index' with no format guidance, validation rules, or required directory structure hints. 'topK' has a one-line description with no minimum/maximum bounds documented in the description itself.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
No error handling guidance visible. errorReporting is declared as true, but there are no recovery hints in descriptions (e.g., 'If indexing fails due to permission errors, try...', or 'If search returns no results, try...').
searchCodebase lacks pagination documentation. Accepting 'topK' is a start, but there is no mention of how to iterate over large result sets, whether results are sorted by relevance, or what happens if topK > available results.
Tool descriptions do not mention indexing cost, latency expectations, or limitations (e.g., max directory size, supported file types, excluded patterns via .gitignore). Users/LLMs cannot reason about when to call these tools.