Applied Intelligence
Module 10: MCP and Tool Integration

Database and API Integration Servers

The integration landscape

Most agentic workflows need external data. Agents query databases, read issues from project trackers, pull documentation, and interact with collaboration tools. The MCP ecosystem has production-ready servers for these common integrations.

This page covers the servers that handle the bulk of enterprise integration needs: databases, source control, communication platforms, and productivity tools. Knowing what already exists saves you from building what someone else already built.

Database servers

Database access is the most requested integration category. An agent that can query production data understands context that no amount of documentation conveys.

PostgreSQL

The original Anthropic reference implementation (@modelcontextprotocol/server-postgres) is archived. A SQL injection vulnerability discovered in April 2025 let attackers bypass read-only restrictions via semicolon-delimited statements. The fix landed in git but the vulnerable version remains on npm with 21,000 weekly downloads.

Production alternatives exist. Postgres MCP Pro from CrystalDBA provides configurable read/write access, explain plan analysis, and index tuning recommendations. The Zed fork (@zeddotdev/postgres-context-server) patches the vulnerability for simple read-only use cases.

What PostgreSQL servers can do:

  • List databases and tables
  • Inspect schemas (columns, types, indexes)
  • Execute SQL queries
  • Connect to multiple databases

Do not deploy the archived @modelcontextprotocol/server-postgres in production. The SQL injection vulnerability affects all versions on npm. Use Postgres MCP Pro or the Zed fork instead.

SQLite

The Anthropic SQLite server (@modelcontextprotocol/server-sqlite) is also archived due to SQL injection vulnerabilities. Over 5,000 downstream forks inherited the vulnerability before archival.

Better implementations offer AI-native features: 73 specialized tools for analytics, JSON operations, text processing, vector search, and geospatial queries. WAL mode support enables better concurrent access. Full-text search via FTS5 works out of the box.

SQLite servers remain useful for local development and testing. For production, verify the implementation patches the injection vulnerability.

DBHub

DBHub from Bytebase is a universal database gateway with enterprise features. One server supports PostgreSQL, MySQL, MariaDB, SQL Server, and SQLite.

What it does:

  • Executes SQL with transaction support and safety controls
  • Searches and explores schemas across multiple databases
  • Provides a web interface for query execution and request tracing
  • Handles SSH tunneling and TLS encryption
  • Enforces read-only mode, row limits, and query timeouts

Configuration uses TOML files that specify multiple database connections. The design minimizes context consumption.

DBHub has 105,000+ downloads and is the recommended choice for multi-database environments.

Time-series and specialized databases

GreptimeDB MCP handles observability data: metrics, logs, and events. SQL and TQL (PromQL-compatible) queries work. A three-layer security model includes read-only user support, application-level gates, and automatic data masking.

AWS Aurora DSQL provides distributed SQL access with natural language query conversion. Read-only by default; write access requires the --allow-writes flag.

Choosing a database server

ServerBest forTrade-off
DBHubMulti-database enterprise environmentsRequires TOML configuration
Postgres MCP ProPostgreSQL with query optimizationMore setup than simple read access
Zed forkSimple PostgreSQL read-onlyMinimal features
GreptimeDBTime-series and observabilitySpecialized use case
AWS Aurora DSQLDistributed cloud SQLAWS-specific

GitHub MCP server

The GitHub MCP server is the most downloaded server in the ecosystem at 889,000+ downloads. GitHub maintains it.

What it does

Repository operations let you browse and query code, search files across repositories, analyze commit history, and understand project structure.

Issue and PR management covers creating, updating, and managing issues; reviewing and commenting on pull requests; triaging bugs with label management; and maintaining project boards.

CI/CD integration includes monitoring GitHub Actions workflows, analyzing build failures, managing releases, and tracking deployment pipeline status.

Security features let you examine security findings, review Dependabot alerts, and access code scanning results.

Authentication

Two options:

OAuth 2.1 with PKCE (recommended): One-click setup in VS Code, Visual Studio, JetBrains, Eclipse, Xcode, and Cursor. Supports SAML enforcement, automatic token refresh, and short-lived credentials. Scoped access limits permissions to specific repositories or organizations.

Personal Access Tokens: Manual configuration for environments where OAuth flows are impractical. Organizations can restrict PAT usage through policy settings.

Organization members with Copilot Business or Enterprise plans must enable the "MCP servers in Copilot" policy. OAuth policies may require explicit app authorization per client.

Slack MCP server

Slack's official MCP server is rolling out to select partners with broad availability expected by summer 2025.

What it does

  • Searches message history with date, user, and content filters
  • Sends messages to any conversation type
  • Retrieves full channel histories including threads
  • Creates and shares canvases (formatted documents)
  • Exports canvases as markdown
  • Accesses user profiles and status information
  • Searches public and private channels

Authentication and admin controls

Admin approval is required. Workspace administrators control which MCP clients can connect. All actions respect existing Slack permissions. Users cannot access channels or messages they could not access through the Slack interface.

The admin control panel in Slack's administrative interface manages all MCP integrations.

Atlassian servers (Jira and Confluence)

The Atlassian Rovo MCP Server provides official cloud-hosted access to Jira, Confluence, and Compass.

What it does

  • Searches and summarizes across all Atlassian products
  • Creates and updates Jira issues via natural language
  • Generates Confluence pages from specifications
  • Handles bulk content operations
  • Enables cross-product workflows (meeting notes to tickets to documentation)

Security model

OAuth 2.1 handles authentication. All actions respect the user's existing access controls: project roles, space permissions, and IP allowlisting rules. Audit logging tracks key actions. Administrators manage external tool connections.

Atlassian MCP is Cloud-only. Server and Data Center deployments do not support the official MCP server. Community alternatives exist for self-hosted Atlassian installations.

Notion MCP server

Notion's official hosted server provides full workspace access.

What it does

  • Reads and writes pages as native operations
  • Searches across workspace content
  • Generates documents: PRDs, technical specs, release notes
  • Automates task creation and management
  • Compiles reports from multiple sources

Notion provides a markdown API optimized for AI agents. The format uses fewer tokens than general-purpose API responses.

Connection methods

  • In-app directory of featured AI tools (one-click)
  • Manual connection via public URL: https://mcp.notion.com/mcp
  • Custom MCP client configurations

Google integrations

Google Cloud announced official MCP servers in December 2025.

Official Google Cloud servers

BigQuery enables direct schema interpretation and native query execution. Data remains in place rather than moving into the context window. Forecasting and analytics work through natural language.

Google Maps provides location-based integrations and geocoding.

Google Compute Engine and Google Kubernetes Engine allow infrastructure management through MCP.

Expansion plans include Cloud Storage, Cloud SQL, Cloud Logging, and security services.

Google Workspace

No official MCP servers exist for Google Sheets, Drive, or Docs as of January 2026. Community implementations (google_workspace_mcp, google-workspace-mcp) provide Gmail, Calendar, Drive, Docs, and Sheets access.

Community servers vary in security implementation. OAuth 2.0/2.1 flows are standard, but review the specific implementation before production use.

When to use existing servers

The general guidance: exhaust existing options before building custom servers.

Use existing servers when:

  • The functionality matches common patterns (GitHub, Slack, databases)
  • Time-to-deployment matters
  • Standard security compliance is sufficient
  • Multiple team members need the same integration

Build custom when:

  • Proprietary data sources have no existing coverage
  • Industry-specific workflows require specialized logic
  • Security requirements exceed off-the-shelf solutions
  • Legacy systems need cohesive integration
  • Custom error handling or agent decision logic is required

Check the MCP Registry at registry.modelcontextprotocol.io before building. Third-party directories like mcp.so and Glama.ai index over 17,000 implementations. The server you need may already exist.

A hybrid approach works well: use existing servers for standard integrations (Slack, GitHub, databases) and build custom servers for sensitive or organization-specific systems.

Security patterns across API servers

All major API integration servers share security patterns:

  • OAuth 2.1 with PKCE is standard for remote authentication
  • Permission inheritance: MCP respects existing platform permissions
  • Audit logging: Tracked at the platform level
  • Admin controls: Management interfaces for organizational oversight
  • Encrypted transport: HTTPS/TLS 1.2+ required

The archived Anthropic database servers demonstrate that even reference implementations can have vulnerabilities. Review any server's security posture before production deployment. Prefer actively maintained implementations over archived reference servers.

On this page