A unified MCP (Model Context Protocol) proxy server that discovers, aggregates, and manages multiple MCP servers with tool discovery, multi-server proxying, and authentication support
This MCP server exhibits significant quality issues across naming, descriptions, schemas, and error handling. Tool definitions are partially visible but lack systematic detail. The six tools have inconsistent quality: some lack actionable descriptions, parameters are minimally documented, and output schemas are not explicitly defined in the visible source code. This is a community-grade server with gaps typical of early-stage implementations.
Get detailed information about a specific cluster
Get detailed information about a specific incident by its number
Simple greeting tool that takes a name and returns a greeting message
List all Databricks clusters with their status and configuration
List all Databricks jobs
Search for ServiceNow incidents by various criteria (priority, state, category, assignment group, correlation_id)
No output schemas documented for any tool. LLMs cannot predict response structure, forcing them to guess field names and types. This violates the tool-chain pattern, if search_incidents returns incident_id and get_incident accepts incident_number, the mismatch is undocumented and forces the LLM to reason about field mapping.
Descriptions are generic and lack decision-making context. 'List all Databricks clusters' doesn't explain: What is this used for? When should I call this instead of get_cluster? What fields does it return? LLMs cannot prioritize tool selection with such vague guidance.
Parameter descriptions lack format constraints and validation rules. For example, state_filter accepts values like 'RUNNING, TERMINATED, PENDING' but is not declared as an enum. Priority and state in search_incidents use numeric codes (1=Critical, 2=High) but LLMs cannot reliably produce correct codes without enums. Free-form strings invite hallucinated values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 60 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 49 | 2024-11-05+ | v1 |
No pagination or result limiting behavior documented. list_clusters and list_jobs could return hundreds of items, but no mention of page/offset/limit parameters, max result count, or continuation tokens. This risks LLMs receiving massive responses that consume context.
Error handling is absent. No guidance on what happens if cluster_id doesn't exist, if search returns no results, or if API credentials are invalid. LLMs receive bare error codes instead of recovery steps like 'Try a different search query' or 'Use list_clusters to find valid IDs'.
Parameter naming inconsistency: search_incidents uses incident_number but get_incident also uses incident_number, good. However, list_jobs has name_filter but search_incidents has a 'query' parameter. The semantic difference is unclear to LLMs: does name_filter search only the job name field, while query searches all fields? This ambiguity forces the LLM to experiment.