ADR-018: Server Functionality Architecture

Date: August 17, 2025 Status: Accepted Deciders: Principal Architect, Chief Architect, Chief of Staff Classification: Architectural (High Impact)

Context

Piper Morgan was originally designed as an MCP consumer only (ADR-001), focusing on consuming external services to enhance PM capabilities. Through PM-033’s strategic expansion, we’ve achieved a complete transformation:

  1. PM-033c (August 14, 2025): Successfully implemented dual-mode MCP server, exposing SpatialIntentClassifier and QueryRouter as MCP services with 0.55-7.18ms latency.

  2. PM-033d (August 16, 2025): Enhanced autonomy with multi-agent coordination achieving 0ms coordination latency (1000x performance target exceeded), 4+ hours autonomous operation demonstrated.

  3. ADR-017: Consolidated spatial intelligence into MCP services, proving the architectural pattern.

This ADR formalizes the COMPLETED server transformation, documenting how Piper has already evolved from tool to platform. We’re capturing the architectural significance of this achieved implementation, not proposing future work.

Decision

We formalize the COMPLETED MCP server transformation achieved through PM-033c and PM-033d, recognizing Piper Morgan’s evolution from standalone tool to composable infrastructure for AI-driven project management.

Achieved Implementation

PM-033c Foundation (August 14):

PM-033d Enhancement (August 16):

This ADR Documents:

Core Capabilities to Expose

  1. PM Intelligence Services:
    • Intent classification and routing
    • Workflow orchestration patterns
    • Project context understanding
    • Ambiguity assessment (ADR-016)
  2. Spatial Intelligence Services (already implemented):
    • 8-dimensional project analysis
    • Pattern recognition across repositories
    • Relationship mapping between entities
    • Temporal pattern detection
  3. Verification Services:
    • Wild claim validation (ADR-015)
    • Metric verification protocols
    • Confidence scoring systems
  4. Orchestration Services:
    • Chain-of-Draft optimization
    • Multi-agent coordination patterns
    • Token-efficient debate protocols

Technical Architecture

External Agent
     ↓
MCP Protocol Request
     ↓
Piper MCP Server Layer (NEW)
├── Authentication & Rate Limiting
├── Capability Discovery
├── Service Router
└── Response Formatting
     ↓
Piper Core Services (EXISTING)
├── Domain Intelligence
├── Workflow Engine
├── Spatial Analysis
└── Pattern Recognition
     ↓
MCP Protocol Response

Implementation Approach

Phase 1: Foundation ✅ COMPLETE (PM-033c)

Phase 2: Enhancement ✅ COMPLETE (PM-033d)

Phase 3: Ecosystem (Next Steps)

Consequences

Positive

  1. Platform Economics: Transform from single-use tool to multi-agent infrastructure
  2. Network Effects: Each new consumer increases Piper’s value
  3. Ecosystem Leadership: Establish Piper as reference PM intelligence provider
  4. Revenue Potential: Enterprise teams could subscribe to Piper’s services
  5. Learning Loops: Usage patterns from other agents improve Piper
  6. Composability: Other tools can build on Piper’s intelligence

Negative

  1. Complexity Increase: Server infrastructure adds operational overhead
  2. Security Burden: Exposing services requires robust security
  3. API Stability: Breaking changes affect external consumers
  4. Support Load: External developers need documentation and help
  5. Performance Impact: Serving others may affect core operations

Neutral

  1. Versioning Strategy: Need semantic versioning for external APIs
  2. Resource Allocation: Balance between self-use and serving others
  3. Monitoring Requirements: Track usage across multiple consumers
  4. Documentation Needs: Comprehensive API documentation required

Alternatives Considered

Alternative 1: Remain Consumer-Only

Why Rejected: Misses strategic opportunity revealed by ADR-017’s success. The ease of exposing spatial intelligence as MCP services proved the pattern’s viability.

Alternative 2: Custom API Instead of MCP

Why Rejected: Would fragment the ecosystem. MCP provides standardization that enables interoperability with other AI agents.

Alternative 3: Delayed Implementation

Why Rejected: First-mover advantage in PM intelligence services is valuable. The foundation from ADR-017 makes this a natural next step.

Implementation Details

Security Model

class MCPServerSecurity:
    """Security layer for MCP server exposure."""

    authentication_methods = [
        "api_key",      # Simple for development
        "oauth2",       # Enterprise integration
        "mcp_native"    # Protocol-native auth
    ]

    rate_limits = {
        "free_tier": "100 requests/hour",
        "standard": "1000 requests/hour",
        "enterprise": "unlimited with fair use"
    }

    capability_tiers = {
        "basic": ["intent_classification", "workflow_routing"],
        "advanced": ["spatial_analysis", "orchestration"],
        "premium": ["full_access", "custom_models"]
    }

Service Discovery

@mcp.list_tools()
async def list_pm_capabilities():
    """Expose available PM intelligence services."""
    return [
        {
            "name": "classify_intent",
            "description": "Classify PM-related intent with 94.7% accuracy",
            "parameters": {...}
        },
        {
            "name": "analyze_project_health",
            "description": "8-dimensional project health analysis",
            "parameters": {...}
        },
        {
            "name": "orchestrate_workflow",
            "description": "Generate optimal workflow for PM tasks",
            "parameters": {...}
        }
    ]

Performance Considerations

Metrics and Success Criteria

Achieved Metrics ✅

Future Success Metrics

Migration Strategy

  1. Internal Dogfooding: Piper consumes its own MCP services
  2. Alpha Partners: 3-5 trusted teams for early feedback
  3. Beta Release: Public beta with rate limits
  4. General Availability: Full launch with SLAs

Notes

The transformation from consumer to provider is not theoretical—it’s operational. PM-033c and PM-033d demonstrated unprecedented success:

This ADR documents a realized vision. What started as an MCP consumer experiment has evolved into a functioning platform providing intelligence services to other agents. The enhanced autonomy breakthrough validates both the technical architecture and the systematic methodology that enabled it.

The success metrics aren’t aspirational—they’re achieved. The platform transformation isn’t planned—it’s complete. This decision captures the architectural significance of what has already been built.

Future Considerations

“From tool to platform, from consumer to provider, from isolated intelligence to ecosystem infrastructure.”