Local-first MCP server for context compilation — deterministic context compilation pipeline with SQLite-backed storage for Cursor, Claude Code, and other MCP clients
AIC provides 7 tools with consistent naming (verb_noun pattern: aic_compile, aic_status, etc.) and reasonable descriptions (avg ~150 chars). All tools are READ_ONLY and well-scoped. However, input schemas lack critical details: no enum constraints, no min/max bounds on integer parameters (timeRangeDays, windowDays), and parameter descriptions are minimal (avg ~40 chars). Output schemas are not documented in the source, only metadata hints in descriptions. Error handling is absent; no guidance on what to do if compilation fails or projectRoot is invalid. Tool composition is sound (each does one thing), but descriptions could be more LLM-optimized per the 50-200 char baseline.
Compiles project context into an optimized prompt for the current task. Returns compiled prompt, metadata (token counts, reduction percentage, task classification, cache hit status, duration), conversation ID, and optional update messages or warnings.
Compiles TypeScript type specifications into optimized documentation. Returns compiled specification and metadata including token counts, reduction percentage, and type tier information.
Generates a summary of the current conversation including task classification, files involved, and compilation statistics.
Inspects the compilation pipeline for a given intent and project, returning detailed trace information about file selection, token allocation, and transformation steps.
Tests the model detection and configuration system for a given project, returning detected model information and configuration state.
Generates a quality report for recent compilations including token efficiency, cache hit rates, task classification accuracy, and guard findings.
Input schemas lack constraints: timeRangeDays and windowDays are integers with no min/max bounds, allowing invalid values (negative, zero, or absurdly large). LLMs will pass unconstrained values.
Output schemas are not documented. Descriptions mention 'returns compiled prompt, metadata (token counts, reduction percentage, ...)' but the actual JSON structure is not specified. LLMs cannot plan downstream operations without knowing field names and types.
No error handling guidance. If projectRoot is invalid or compilation fails, there is no documented recovery path. Descriptions do not explain what errors are possible or how to handle them.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2025-06-18+ | v2 |
Returns current AIC server status including version, installation scope, editor detection, configuration state, and recent compilation statistics.
Parameter descriptions are minimal (avg ~40 chars, below 72-char baseline). 'Absolute path to the project root directory' is clear, but 'Number of days to include in status report (default 7)' lacks constraint info (e.g., 'must be 1 - 365').
Tool descriptions do not explain WHEN to use each tool or how they differ. aic_status, aic_quality_report, and aic_conversation_summary all return compilation statistics, the distinctions are unclear to an LLM.