MCP server for Java/multi-language code graph analysis, stack trace debugging, and documentation retrieval
JIDRA provides 9 tools with complete input schemas and descriptions, but quality is uneven. All tools have descriptions (10-200 chars, baseline 194), and all parameters are typed with descriptions. However, descriptions are terse and lack LLM-optimization guidance (e.g., 'when to use', 'prerequisites', 'what it returns'). No output schemas are documented, LLMs cannot plan downstream calls or extract structured data. Error handling is absent from descriptions. Tools like jidra_reindex and jidra_analyze_stack_trace combine multiple concerns (e.g., reindex + optional file hints; analyze + flow depth + node limits), violating single-responsibility. Parameter defaults are reasonable but undocumented. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY vs WRITE). Naming is verb-first and clear (search_, find_, get_, analyze_), but some names are domain-specific jargon ('jidra_' prefix, 'graph_path', 'method_id') that may confuse agents unfamiliar with code graph terminology.
Analyze a stack trace against the code graph to identify failure points, suggest debug locations, and provide focused flow maps.
Check whether the graph is stale — call this at session start.
Find all callees (methods called by) a given method in the code graph.
Find all callers of a given method in the code graph.
Retrieve documentation chunks linked to a class or matching a free-text query.
Resolved/unresolved/external callsite breakdown for the local code graph.
Reindex the codebase incrementally.
No output schemas documented. LLMs cannot infer what fields to expect from tool responses, preventing downstream tool chaining and forcing agents to guess at response structure.
Descriptions lack LLM-optimization guidance. Missing 'WHEN to use', 'prerequisites', and 'what it returns'. Baseline is 194 chars; most here are 30-100 chars with minimal context.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite explicit risk levels. jidra_reindex is WRITE but unmarked; others are READ_ONLY but unmarked. Agents cannot infer safety/retry semantics.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Search for classes by name or full name in the code graph.
Search for methods by name or signature in the code graph.
jidra_analyze_stack_trace combines multiple concerns: stack trace analysis + flow depth control + node limits + utility filtering. Should split into separate tools or document parameter dependencies.
No error handling guidance in descriptions. Tools do not document what happens on failure (e.g., graph not found, method_id invalid, stack trace malformed) or how agents should recover.