The MCP Ecosystem
Relevant source files
The following files were used as context for generating this wiki page:
The Model Context Protocol (MCP) ecosystem within this project serves as a unified orchestration layer designed to solve universal AI challenges such as context drift, confidence scoring, and multi-step research. It acts as a bridge between AI agents (like Claude) and project-specific tools, databases, and research engines to enhance reliability and output validation.
The ecosystem is categorized into core AI enhancement servers, database connectors, and specialized research engines. These components work together to provide a "deterministic gate" for AI operations, ensuring that generated outputs meet engineering standards through automated validation and constraint checking.
Sources: README.md:12-25, mcp-servers/README.md:3-8, PROMPT-ADD-TO-EXISTING.md:120-135
Core Architecture and Data Flow
The ecosystem implements a sophisticated research pipeline known as the ARE/QRE (Adaptive Research Engine / Query Refinement Engine). This pipeline orchestrates requests through multiple layers of analysis, validation, and scoring before final report generation.
ARE/QRE Pipeline Flow
The following diagram illustrates how a research request progresses through the specialized MCP servers.
The ecosystem utilizes specific ports and specialized logic for each stage, moving from discovery to validation and synthesis.
Sources: mcp-servers/README.md:12-42
MCP Server Catalog
The platform provides a suite of domain-agnostic servers that address technical, financial, or medical research needs by providing structured interfaces to AI models.
AI Research & Analysis Servers
| Server | Port | Key Capabilities | Use Cases |
|---|---|---|---|
| research-swarm | 3012 | Multi-mode orchestration, formula scoring, pattern detection, circuit breaking. | Academic review, financial due diligence. |
| basin-analyzer | 9383 | Context drift detection via Epsilon (ε) metrics and basin clustering. | AI quality control, prompt A/B testing. |
| report-writer | 9386 | Quality assessment, structured formatting, citation validation. | Technical docs, executive summaries. |
| constraint-validator | 9385 | Z3 SMT solver for logical conflict detection. | Business rule validation, data consistency. |
Sources: mcp-servers/README.md:46-155, README.md:104-113
Specialized Infrastructure & Data Servers
- ragflow (Port 3010): Orchestrates RAG with hybrid search (keyword + semantic) and multi-KB federation using Elasticsearch.
- smart-memory: Manages cross-platform team knowledge via a knowledge graph (entities, relations, and observations).
- team-analytics: Tracks engineer health scores (0-100) based on code quality, velocity, and collaboration metrics.
- Database Connectors: Generic servers for PostgreSQL (Port 9384) and MySQL (Port 9385) providing schema introspection and sanitized query execution.
Sources: mcp-servers/README.md:157-200, mcp-servers/README-SETUP.md:65-71, mcp-servers/team-analytics/README.md:8-25
Operational Management
The ecosystem includes strict evaluation and integration protocols to maintain high quality standards and prevent "context window bloat."
Server Evaluation Framework
New servers are evaluated against a 100-point rubric across four categories to determine their "Tier" status.
Sources: mcp/evaluation-criteria.md:7-40
Integration Configuration
Servers are typically configured via a .mcp.json file in the project root or within AI client settings (e.g., Claude Code or Cursor).
{
"mcpServers": {
"beads-integration": {
"command": "node",
"args": ["${workspaceFolder}/.eng-platform/mcp-servers/beads-integration/server.js"]
},
"basin-analyzer": {
"command": "python",
"args": ["${workspaceFolder}/.eng-platform/mcp-servers/basin-analyzer/server.py"],
"env": {
"VOYAGE_API_KEY": "${VOYAGE_API_KEY}",
"DEEPSEEK_API_KEY": "${DEEPSEEK_API_KEY}"
}
}
}
}
Sources: PROMPT-ADD-TO-EXISTING.md:46-75, mcp-servers/README.md:213-233
Key Metrics and Performance
The ecosystem prioritizes throughput and stability. For example, the research-swarm server is benchmarked to handle 50 saves per second with a p95 latency under 200ms. The basin-analyzer utilizes an Epsilon (ε) metric (0 to 1) where lower scores indicate higher confidence and output stability across LLM response patterns.
Sources: mcp-servers/README.md:83-93, mcp-servers/README.md:246-260
Conclusion
The MCP Ecosystem provides the infrastructure for AI-native engineering. By combining research orchestration (research-swarm), drift detection (basin-analyzer), and team health monitoring (team-analytics), the platform transforms LLMs from simple chat interfaces into structured, validated engineering tools. This system ensures that as AI agents interact with the codebase, their actions remain consistent with logical constraints and project context.
Sources: README.md:7-10, mcp-servers/README.md:280-285