An MCP-based agentic orchestrator that coordinates multiple AI agents (Supervisor, Researcher, Analyst) to perform research, analysis, and report generation using web search, scraping, and vector database storage.
Server has 10 tools with basic schemas and descriptions, but significant quality gaps. Tool naming is inconsistent (duplicate tools with different prefixes: duckduckgo_search appears twice, store_research appears twice with different schemas, retrieve_knowledge appears twice). Descriptions are present but generic (avg ~60 chars, below 194-char baseline). Parameters lack depth, no enums for constrained inputs, no format specifications, no validation guidance. Output schemas are undocumented (tools return str or List[str] with no field documentation). Error handling is absent, no recovery guidance, no categorization. Security concerns: no mention of API key injection, no audit logging visible. Composition issues: duplicate tools suggest incomplete refactoring.
Search the web using DuckDuckGo.
Search the web using DuckDuckGo.
Retrieve semantically similar research entries from the vector database.
Embed and store research text in the vector database.
Fetch a URL and return the first 5000 characters of page text.
Retrieve semantically similar research entries with source URLs.
Retrieve semantically similar research entries from the vector database.
Duplicate tool definitions with inconsistent naming. Tools appear multiple times (duckduckgo_search, store_research, retrieve_knowledge) with different prefixes (local_mcp_*) and sometimes different schemas. This creates ambiguity for LLMs and suggests incomplete refactoring.
Output schemas are completely undocumented. Tools return str or List[str] with no field documentation. LLMs cannot plan downstream calls or extract structured data. E.g., retrieve_knowledge returns List[str] but what fields does each entry contain? Does it include source_url?
Descriptions are generic and lack LLM-optimization guidance. Average ~60 chars (baseline 194). Missing WHEN to use, prerequisites, and dependencies. E.g., 'Embed and store research text' doesn't explain when to call this vs retrieve_knowledge, or what happens if the vector DB is unavailable.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
Embed and store research text in the vector database.
Embed and store research text in the vector database.
Fetch a URL and return the first 5000 characters of page text.
No error handling or recovery guidance. Tools lack try-catch blocks visible in source, no error categorization (retryable vs fatal), and no actionable error messages. If web_scraper fails, the LLM gets no guidance on what to do next.
Parameters lack constraints and validation guidance. max_results and k have no min/max bounds documented. No enums for constrained inputs. No format specifications (e.g., URL format for web_scraper). LLMs can pass invalid values without guidance.