Skip to main content
Glama
Combined MCP Server

Combined MCP Server

by manish6007

Combined MCP Server

A production-grade MCP (Model Context Protocol) server combining Redshift query capabilities and Knowledgebase vector store features.

Features

Redshift Tools

  • run_query - Execute SQL with IAM authentication via get_cluster_credentials

  • list_schemas - List database schemas

  • list_tables - List tables in a schema

  • describe_table - Get table structure

Large results (>100 rows) are automatically stored in S3 with 20 sample rows returned.

Knowledgebase Tools

  • build_vectorstore - Build vector store from S3 markdown files

  • query_vectorstore - Hybrid search (semantic + keyword) with RRF reranking

  • get_vectorstore_status - Check build status and cache stats

Related MCP server: Redshift MCP Server

Quick Start

Local Development

  1. Install uv (if not already installed):

    curl -LsSf https://astral.sh/uv/install.sh | sh
    # Or on Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
  2. Start infrastructure:

    docker-compose up -d postgres localstack
  3. Install dependencies:

    uv pip install -e ".[dev]"
  4. Configure environment:

    cp .env.example .env.local
    # Edit .env.local with your settings
  5. Run the server:

    # With MCP Inspector
    mcp dev src/combined_mcp_server/main.py
    
    # Or directly
    python -m combined_mcp_server.main

ECS Deployment

# Build container
docker build -t combined-mcp-server .

# Run with health checks
docker run -p 8080:8080 --env-file .env combined-mcp-server

Health endpoints:

  • GET /health - Liveness probe

  • GET /ready - Readiness probe

  • GET /status - Detailed status

Configuration

See .env.example for all configuration options. Key settings:

Variable

Description

REDSHIFT_CLUSTER_ID

Redshift cluster identifier

POSTGRES_SECRET_NAME

Secrets Manager secret for pgvector DB

KNOWLEDGEBASE_S3_BUCKET

S3 bucket with markdown files

BEDROCK_EMBEDDING_MODEL

Titan embedding model ID

Architecture

┌─────────────────────────────────────────────────────┐
│                  Combined MCP Server                 │
├─────────────────────┬───────────────────────────────┤
│   Redshift Tools    │     Knowledgebase Tools       │
│  ─────────────────  │  ───────────────────────────  │
│  • run_query        │  • build_vectorstore          │
│  • list_schemas     │  • query_vectorstore          │
│  • list_tables      │  • get_vectorstore_status     │
│  • describe_table   │                               │
├─────────────────────┴───────────────────────────────┤
│                    Core Services                     │
│  AWS (Secrets Manager, S3, Bedrock, Redshift)       │
│  PostgreSQL + pgvector                              │
└─────────────────────────────────────────────────────┘

Testing

# Unit tests
pytest tests/ -v

# With coverage
pytest tests/ -v --cov=combined_mcp_server

# Integration tests (requires Docker)
docker-compose up -d
pytest tests/ -v -m integration

License

MIT

Available Tools

7 tools
build_vectorstoreA
Build or rebuild the knowledge base vector store from S3 markdown files.

Downloads all markdown files from the configured S3 location, processes them 
into chunks, generates embeddings using AWS Bedrock Titan, and stores in PostgreSQL.

Inputs: no named parameters. Returns: object

describe_tableB
Get detailed information about a Redshift table including columns and data types.

Args:
    schema: Schema name
    table: Table name
    db_user: Database user for authentication
    db_group: Optional database group for permissions

Inputs: schema (required), table (required), db_user (required), db_group. Returns: object

get_vectorstore_statusA
Get the current status of the knowledge base vector store.

Returns build status, document count, and cache statistics.

Inputs: no named parameters. Returns: object

list_schemasB
List all schemas in the Redshift database.

Args:
    db_user: Database user for authentication
    db_group: Optional database group for permissions

Inputs: db_user (required), db_group. Returns: object

list_tablesB
List all tables in a Redshift schema.

Args:
    schema: Schema name to list tables from
    db_user: Database user for authentication
    db_group: Optional database group for permissions

Inputs: schema (required), db_user (required), db_group. Returns: object

query_vectorstoreA
Search the knowledge base vector store.

Supports semantic search (vector similarity), keyword search (full-text), 
or hybrid search combining both with RRF reranking. Results are cached for performance.

Args:
    query: The search query text
    top_k: Maximum number of results to return (default: 10)
    search_type: Type of search - semantic, keyword, or hybrid (default)

Inputs: query (required), top_k, search_type. Returns: object

run_queryA
Execute a SQL query on Redshift.

For queries returning more than 100 rows, the full result set is stored 
in S3 and only 20 sample rows are returned.

Args:
    sql: The SQL query to execute
    db_user: Database user for authentication via get_cluster_credentials
    db_group: Optional database group for permissions

Inputs: sql (required), db_user (required), db_group. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose with no overlap. The vectorstore tools (build, get_status, query) handle knowledge base operations, while the Redshift tools (list_schemas, list_tables, describe_table, run_query) handle database operations. The separation between these two domains is unambiguous.

    Naming Consistency5/5

    All tools follow a consistent verb_noun naming pattern. The verbs are descriptive (build, describe, get, list, query, run) and the nouns clearly indicate the target resource (vectorstore, table, schemas, tables, query). There are no deviations in naming conventions.

    Tool Count4/5

    With 7 tools, the count is reasonable for a combined server covering two domains. However, the vectorstore side feels slightly thin with only three tools (build, status, query), lacking operations like update or delete. The Redshift side is well-covered with four tools.

    Completeness4/5

    For the Redshift domain, the toolset provides comprehensive coverage (list schemas/tables, describe table, run query). For the vectorstore domain, core operations exist (build, query, status), but there are minor gaps such as updating or deleting documents from the knowledge base, which agents might need to work around.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers