Transform diagrams, screenshots, and rough ideas into polished Notion artifacts with Gemini and Notion-ready publishing.
NexusForge exposes 7 HTTP API endpoints as MCP tools, but lacks critical definition quality. Tool names mix verbs inconsistently (health_check vs nexus_generate_and_publish). Most descriptions are present but generic (50-100 chars, below the 194-char baseline). Input schemas are partially visible for 2 tools (nexus_generate_and_publish, notion_oauth_callback) but missing entirely for 5 tools (health_check, notion_oauth_connect, notion_disconnect, notion_search_pages, notion_get_session). Parameters lack type definitions in visible schemas. Output schemas are undocumented. Error handling is minimal, no recovery guidance or error classification. The nexus_generate_and_publish tool combines content generation AND optional publishing, violating single-responsibility. No tool annotations (readOnlyHint/destructiveHint). Security concern: OAuth tokens stored in cookies, not validated as server-side secrets.
Check health status of Gemini configuration, Notion connection, and OAuth setup
Generate content using Gemini based on prompt and optional image, then optionally publish to Notion
Clear Notion OAuth session cookie and disconnect Notion
Get current Notion OAuth session information
Handle Notion OAuth callback and exchange authorization code for access token
Initiate Notion OAuth authorization flow
Search Notion pages by query string
5 of 7 tools have NO input schema visible in source code.
nexus_generate_and_publish combines two responsibilities: content generation AND optional Notion publishing. Should split into separate tools (generate_content, publish_to_notion) so agent can compose independently.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., notionPageId from search results).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | 2026-07-28+ | v2 |
Descriptions are 50-65 chars (well below 194-char baseline). Too brief to guide LLM selection. E.g., 'Check health status...' lacks WHEN to call it or what prerequisites exist.
No error handling guidance. Errors return raw HTTP status codes (400, 500, 502) with minimal context. LLM cannot determine if error is retryable, user-fixable, or fatal.