MCP server that gives AI assistants like Claude access to LinkedIn profiles, companies, and job postings through the user's own browser session.
Single tool 'close_session' with empty input schema {} and minimal description. No parameter documentation, no output schema, no error handling guidance. Tool name lacks clarity on what 'session' means (browser? LinkedIn? auth?). Destructive operation with no confirmation pattern. Does not meet baseline standards for production tools.
Close the current browser session and clean up resources.
Input schema is empty ({}). No parameters documented, no types, no descriptions.
Tool description is 60 characters but lacks context: What session? Browser? LinkedIn auth? When should LLM call it? What does it return? Does not meet 10-1024 char baseline or LLM-optimization standard.
Destructive operation (closes session, cleans resources) with no confirmation pattern, dry-run option, or recovery guidance. Agents will invoke without understanding consequences.
No output schema documented. LLM cannot infer what fields to expect (success boolean? error message? cleanup details?). Breaks downstream tool chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | <=2025-11-25 | v2 |
No error handling guidance. If session close fails, what should LLM do? Retry? Ask user? Fatal? No recovery path provided.