Glama
Reaper MCP Server

# Technical Architecture

## Data Flow Architecture

```mermaid
flowchart TB
 subgraph Clients
  WC[WebSocket]
  RC[REST]
  GC[GraphQL]
 end

 subgraph API
  FA[FastAPI]
  WH[WS Handler]
  RH[REST Handler]
  GH[GQL Handler]
  MM[Message Manager]
  
  FA --> WH
  FA --> RH
  FA --> GH
  WH --> MM
  RH --> MM
  GH --> MM
 end

 subgraph Core
  MCP[MCP Service]
  TM[Tool Manager]
  RM[Resource Manager]
  PM[Prompt Manager]
  TR[Tool Registry]
  RC2[Resource Cache]
  PL[Prompt Loader]
  
  MM --> MCP
  MCP --> TM
  MCP --> RM
  MCP --> PM
  TM --> TR
  RM --> RC2
  PM --> PL
 end

 subgraph Storage
  ES[(Elastic)]
  RD[(Redis)]
  
  subgraph ESI[ES Indices]
   VI[Vector]
   FI[Full-text]
   MI[Metadata]
  end
  
  subgraph RDS[Redis Store]
   CS[Cache]
   SS[Session]
   PS[PubSub]
  end
 end

 subgraph Monitor
  PE[Prometheus]
  PA[Alerts]
  GD[Grafana]
  
  PE --> PA
  PA --> GD
 end

 WC --> FA
 RC --> FA
 GC --> FA
 
 MCP --> ES
 MCP --> RD
 
 FA --> PE
```

## Component Details

### API Gateway

- **FastAPI Server**: Main entry point handling all incoming requests
- **WebSocket Handler**: Manages real-time bidirectional communication
- **REST Handler**: Processes HTTP requests for REST API
- **GraphQL Handler**: Handles GraphQL queries and mutations

### MCP Core

- **MCP Service**: Core service implementing Model Context Protocol
- **Tool Manager**: Handles tool registration and execution
- **Resource Manager**: Manages resource loading and caching
- **Prompt Manager**: Processes and manages system/user prompts

### Storage Layer

- **Elasticsearch**:
  - Vector Index: For semantic search
  - Full-text Index: For text search
  - Metadata Index: For storing relationships
- **Redis**:
  - Cache Store: For fast data access
  - Session Store: For user sessions
  - Pub/Sub: For real-time updates

### Monitoring

- **Prometheus**: Metrics collection and alerting
- **Grafana**: Visualization and dashboards

## Data Flow Sequence

1. Client sends request through WebSocket/REST/GraphQL
2. FastAPI routes request to appropriate handler
3. Message Manager processes and validates request
4. MCP Service executes requested operation
5. Results are stored in Elasticsearch/Redis
6. Response is sent back to client
7. Metrics are collected by Prometheus