Advanced MCP server for Storybook with AST parsing, addon management, test generation, and interactive editing
This server has significant quality gaps that would not pass production code review. While all 25 tools have names and basic descriptions, the implementation exhibits several critical deficiencies: (1) Most tool descriptions are generic and lack actionable context for LLM selection (e.g., 'Parses a story file and extracts metadata' tells the LLM when to use parse_story vs parse_component). (2) Parameter descriptions are minimal or missing constraint details (e.g., 'Path to the story file' provides no guidance on absolute vs relative paths, despite the code checking for absolute paths). (3) Output schemas are completely undocumented, tools return JSON blobs without declared field structures, forcing LLMs to parse unstructured responses. (4) Error handling is minimal (basic try-catch returning JSON error objects) with no actionable recovery guidance. (5) Many tools appear to be inferred from test scripts rather than explicitly defined with full schemas (e.g., generate_story_from_component, parse_component exist in test-server.sh but not visible in source index.ts with complete schema definitions). (6) Several tools combine multiple concerns without clear boundaries (e.g., convert_story_to_component does parsing + conversion + optional file write). This is a domain-specific toolkit with incomplete polish.
Adds a Storybook addon to the project configuration
Adds a new story to an existing story file
Batch update multiple args across multiple stories
Clears the MCP server cache
Clones an existing story with optional modifications
Compares component props with story args to identify sync issues
Converts a Storybook story to a standalone React component
Extracts props and args from a story file
Output schemas completely undocumented. Tools return JSON.stringify() wrapped in generic {type:'text', text:JSON} content without declaring what fields the response object contains. LLMs cannot plan downstream tool calls without knowing available fields.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 38 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 29 | - | v1 |
Finds and replaces text patterns in files
Generates accessibility tests for a story
Generates a decorator for reusable story configuration
Generates interaction tests for a story
Generates a play function for interactive testing of a story
Generates a Storybook story file from a React component
Generates unit tests for a story
Gets cache statistics for the server
Gets a list of recommended Storybook addons based on project setup
Lists all configured Storybook addons in a project
Lists all Storybook story files in a directory
Parses a React component and extracts prop information
Parses a story file and extracts metadata
Removes a Storybook addon from the project configuration
Syncs component props based on story arg definitions
Syncs story args to match component prop definitions
Updates specific arg values in a story
Descriptions lack actionable context for LLM tool selection. Generic descriptions like 'Parses a story file and extracts metadata' (parse_story) do not differentiate from parse_component or explain when to call each. LLMs cannot distinguish between overlapping tools without more specific guidance.
Parameter descriptions missing constraint details and format guidance. Examples: 'Path to the story file' provides no indication of absolute vs relative path expectations; 'Name of the story' does not clarify case sensitivity or naming conventions; 'Object with arg names as keys and new values' (argUpdates) lacks type specifications for values.
Tool definitions appear inferred from test scripts rather than explicitly visible in source code. Tools like generate_story_from_component, parse_component exist in test-server.sh but complete schema definitions not visible in index.ts excerpt. This violates the requirement to verify tool registration in source code.
Error handling provides no recovery guidance. try-catch blocks return JSON error objects but do not indicate whether errors are retryable, user-fixable, or fatal. No actionable next steps for LLM. Example: 'Directory not found' returns error without suggesting search_users() or retry logic.
Multiple tools combine concerns without clear boundaries. find_and_replace performs search + replace in a single operation; convert_story_to_component does parsing + conversion + optional write; sync_story_to_component and sync_component_to_story are unclear inverses. This violates single-responsibility principle and makes composition difficult.
Lack of output result limits and pagination for tools that return collections. list_stories returns all matching files without pagination or a max limit parameter. If a project has thousands of story files, the response could exhaust context windows. No cursor, limit, or offset parameters.
No tool annotations for destructiveness or readonly status, despite tools performing writes. Server initialized with McpServer but no tool registration includes readOnlyHint, destructiveHint, or idempotentHint. This deprives agents of static knowledge about which tools are safe to retry or compose.