A browser-native MCP server that exposes Konect4AI research capabilities through WebMCP and A2A (Agent-to-Agent) protocols. Bridges Konect4AI MCP tools to web browsers and external agents.
Konect4AI WebMCP exposes 7 tools via HTTP with basic schemas and descriptions. However, critical gaps undermine production readiness: (1) Most tool descriptions are generic and under 100 chars, lacking WHEN/WHY context for LLM selection. (2) Input schemas present but sparse, many parameters lack type constraints, enums, or validation guidance. (3) No output schemas documented; LLMs cannot predict response structure for chaining. (4) Error handling absent, no recovery guidance, retryability classification, or actionable error messages. (5) Tool composition issues: research_connected_sources, compare_source_findings, summarize_web_data, and analyze_structured_results appear to form a pipeline but lack explicit chaining IDs and dependency documentation. (6) ask_data_source requires jobId (opaque system ID) instead of accepting human-friendly data source names. Baseline: avg tool description 194 chars; this server averages ~80 chars. Avg params per tool 4; this server ranges 1 - 3 with minimal constraints.
Analyze structured results returned by a connected capability and extract patterns.
Ask a question of a connected data source and receive an answer with evidence.
Compare results from multiple sources and surface agreement, disagreement, and gaps.
Search the Data.gov catalog for datasets matching a query.
Propose a new data source to be added to the Konect4AI system.
Research a topic across multiple connected data sources and keep the evidence traceable.
No output schemas documented. LLMs cannot predict response structure, breaking tool chaining and forcing exploratory calls.
Tool descriptions too brief (avg 80 chars vs baseline 194). Missing WHEN/WHY context for LLM selection. E.g., 'Research a topic across multiple connected data sources' does not explain when to use this vs datagov_search or ask_data_source.
ask_data_source requires jobId (opaque system ID) instead of accepting human-friendly data source names. Forces extra lookup call and mismatches chat data model.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2025-06-18+ | v2 |
Turn connected web data into a concise structured summary for review.
Input parameters lack type constraints and validation guidance. E.g., 'sources' array has no item type or format; 'maxSources' has no min/max bounds in schema (only in description). 'data' and 'results' parameters are bare objects with no field documentation.
No error handling guidance. Tools return raw errors with no recovery hints, retryability classification, or actionable next steps for LLM.