Model Context Protocol server for Haiven AI prompts API that allows AI assistants to discover and execute prompts from a Haiven instance
Three tools with mixed definition quality. Tool naming is clear and action-verb-based (get_*, good adherence to pattern:tool). Descriptions are present but lack depth and context around when/why to use each tool. Schemas are properly typed with JSON Schema, but parameter descriptions are sparse. The get_casper_workflow tool is particularly complex with write side effects but lacks error guidance or confirmation patterns. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.
Get or save the Casper workflow methodology as rules/instructions for AI development guidance. Supports two modes: 'share' (returns content to LLM) or 'save' (saves to appropriate tool directory like .cursor/rules or .github/instructions)
Retrieve the full content and metadata of a specific prompt by ID from the Haiven instance
Retrieve a list of all available prompts from the Haiven instance with metadata including title, categories, help text, and description
get_casper_workflow has a WRITE mode (saves files) but lacks confirmation/dry-run pattern and error guidance. No destructiveHint annotation to signal side effects to LLM.
Descriptions are generic and lack context on WHEN to use each tool. Description for get_prompts is 96 chars (below the 194-char production baseline p50). No dependency hints or discovery guidance.
get_casper_workflow parameter descriptions are sparse. 'section' has no explanation of when to request partial vs full. 'tool_context' and 'project_directory' descriptions lack clarity on defaults and auto-detection behavior.
No tool annotations (readOnlyHint for safe tools, destructiveHint for write operations). get_prompts and get_prompt_text marked READ_ONLY but not declared in schema; get_casper_workflow is WRITE but has no annotation.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 48 | - | v1 |
Output schemas are not documented in source code. No structured response format documented for tools, LLM must infer what fields to expect from descriptions alone.
Error handling not visible in tool definitions. No guidance on what happens if Haiven API is unreachable, if prompt_id is invalid, or if file write fails in save mode.