ContextForge MCP server — exposes project context and conventions to AI-powered IDEs via the Model Context Protocol
ContextForge MCP has 4 tools with basic descriptions but significant schema and parameter documentation gaps. Tool names follow verb_noun convention (get_*, log_*), which is good. However, 3 of 4 tools have empty input schemas ({}), violating the critical rule that tools without input schemas must score 0 on schema. The one tool with parameters (get_relevant_context) has minimal description of its single parameter. Output schemas are not documented anywhere in the visible code. Error handling and recovery guidance are absent. The server is STDIO-only, which caps protocol readiness at 50 and limits real-world utility.
Enriches a raw prompt with stack context, conventions, prior decisions, and anti-patterns
Returns all detected project conventions with confidence scores and contested flags
Returns relevant context sections from .context.md for a given feature area
Logs an architectural decision to .contextforge/decisions.jsonl
Three of four tools have empty input schemas ({}). This blocks LLMs from understanding what parameters are required or optional.
Parameter descriptions are minimal or missing. 'feature_area' in get_relevant_context has only 'The feature area or topic to get context for', no guidance on valid values, format, or examples. 'raw_prompt' and decision parameters lack context on expected length, format, or constraints.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. What does get_project_conventions return? A list of objects? A map? What fields? Unknown.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If get_relevant_context receives an invalid feature_area, what happens? No error message tells the LLM what to do next or what valid values are.
Tool descriptions are generic and lack WHEN/WHY guidance. 'Returns all detected project conventions with confidence scores and contested flags' does not explain when to call this vs enrich_prompt, or what the LLM should do with the result.