Graph-powered code intelligence engine for Go. Indexes Go codebases into a structural knowledge graph, enabling powerful code search, impact analysis, and more. Exposes an MCP server interface for integration with Claude and other clients.
Axon Go provides 7 well-named tools with clear READ_ONLY risk classification, but critical gaps in parameter descriptions, output schema documentation, and error handling undermine usability for LLM agents. Tool names follow verb_noun convention well (axon_query, axon_context, axon_impact, axon_dead_code, axon_list_repos, axon_cypher, axon_detect_changes), which is a strength. However, most tools lack detailed parameter descriptions explaining constraints, formats, and expected ranges. Output schemas are not documented, leaving agents unable to plan downstream steps or extract required fields. Error handling is absent, no recovery guidance, no categorization of error types, and no actionable error messages shown in source. The tool descriptions themselves are concise (47-92 chars) but lack WHEN/WHY context that would help LLMs decide between similar tools (e.g., when to call axon_query vs axon_cypher). Input parameter descriptions are either missing entirely (axon_dead_code, axon_list_repos) or generic (e.g., 'Search query text' without format/length constraints). Pagination is not visible for list tools. This server would require significant documentation work before production LLM integration.
Get a 360-degree view of a symbol: callers, callees, type references, and community membership.
Execute a raw Cypher query against the knowledge graph (read-only).
List all symbols detected as dead (unreachable) code.
Detect changes in specified files and analyze their impact on the codebase.
Blast radius analysis: find all symbols affected by changing a given symbol.
List all indexed repositories with their stats.
Search the knowledge graph using hybrid search. Returns ranked symbols matching the query.
Output schemas not documented. No fields are specified for what axon_query, axon_context, axon_impact, or other tools return. LLMs cannot plan multi-step tool chains or extract required IDs for downstream calls without knowing the response structure.
Parameter descriptions missing or insufficient for axon_dead_code and axon_list_repos (empty input {}). For axon_query, 'Search query text' lacks format constraints (length, character restrictions, syntax rules). For axon_impact, 'depth' parameter has no min/max bounds or explanation of semantics.
Tool descriptions lack WHEN/WHY context. For example, axon_query (47 chars) does not explain when to use it vs axon_cypher (which also searches the graph). Descriptions do not state preconditions (e.g., 'Call axon_list_repos first to see available repositories') or what kinds of results to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
No error handling or recovery guidance. Source code does not show error categorization (retryable vs user-fixable vs fatal), actionable error messages, or guidance on next steps when a tool call fails. An LLM hitting an error has no recovery path.
Pagination not visible for list tools (axon_list_repos, axon_dead_code). No limit, offset, page, or cursor parameters shown. If result sets are large, they will blow the LLM context window.
Parameter type constraints missing. 'limit' parameter in axon_query has type 'integer' but no min/max bounds, preventing LLMs from knowing valid ranges (e.g., 1 - 100, 1 - 1000). 'depth' in axon_impact similarly lacks bounds.