Persistent structural knowledge graph for codebases — MCP tools + CLI tree explorer
Codeindex provides 6 well-named, read-only code analysis tools with complete input schemas and clear descriptions. All tools follow verb_noun naming (get_*, find_*, search_*). Descriptions are concise (50-150 chars) and explain purpose clearly. Input schemas are properly typed with required fields. However, output schemas are completely undocumented, LLMs cannot predict response structure or plan downstream calls. Parameter descriptions lack detail on constraints, ranges, and error cases. Error handling is minimal. The reindex tool is a write operation but lacks confirmation/dry-run safeguards. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.
Locate where a function, type, variable, or class is defined across the codebase.
Trace the call graph upstream from a function. Returns caller chain with depth, file paths, and line numbers.
Returns the structural skeleton of a file (symbols, imports) without source code. Includes stale flag.
Find every file and line that uses a given symbol (calls, imports, references).
Retrieve a bounded neighborhood of the knowledge graph around a symbol. Returns nodes and edges within the specified depth.
Trigger re-indexing of a specific file or the full repo.
Output schemas completely undocumented. LLMs cannot predict response structure, field names, or types. Breaks tool chaining and forces agents to guess at result format.
Parameter descriptions lack constraint details. 'depth' parameter has no min/max bounds stated (code shows max 10, but description silent). 'kind' enum values not documented in description. LLMs cannot validate inputs.
reindex tool (write operation) lacks confirmation/dry-run pattern. No error guidance on what happens if reindex fails mid-operation. Agents cannot safely trigger full-repo reindexing.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). 5 tools are read-only, 1 is destructive, annotations would guide agent planning and safety checks.
No pagination support documented. Tools like get_references and get_subgraph could return large result sets. No limit, offset, or cursor parameters visible. Risk of context window exhaustion.