Sign in Subscribe
By Arnav Jalan —

25 Best MCP Servers for AI Agents: Complete Setup Guide (2026)

25 Best MCP Servers for AI Agents: Complete Setup Guide (2026)

Model Context Protocol has become the standard for connecting AI agents to external tools. Since Anthropic donated MCP to the Linux Foundation in December 2025, adoption has accelerated across OpenAI, Google, and most major AI platforms. This shift reflects broader enterprise AI trends toward standardized tooling.

This guide covers 25 MCP servers across six categories: productivity, databases, development tools, browser automation, cloud infrastructure, and search. Each entry includes setup configuration, authentication requirements, key tools exposed, and practical use cases.

How MCP Servers Work

MCP servers expose tools, resources, and prompts to AI clients through a standardized JSON-RPC protocol. The client (Claude Desktop, Cursor, Claude Code) connects to servers via stdio, HTTP, or SSE transport.

A typical configuration in claude_desktop_config.json:

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "@package/server-name"],
      "env": {
        "API_KEY": "your-key"
      }
    }
  }
}

Remote servers use URL-based configuration:

{
  "mcpServers": {
    "remote-server": {
      "url": "https://mcp.service.com/mcp",
      "transport": "sse"
    }
  }
}

Productivity & Communication

1. Slack MCP Server

Maintainer: Slack (official, hosted at mcp.slack.com)
Transport: Remote SSE with OAuth

Slack's official MCP server provides 47 tools for workspace interaction. Connect once via OAuth, and your agent can search messages, post to channels, and manage workflows.

Key Tools:

  • slack_send_message - Post to channels by ID
  • slack_search_public - Search public channel messages
  • slack_search_public_and_private - Search all channels
  • slack_read_channel - Fetch message history
  • slack_read_thread - Get thread replies
  • slack_create_canvas - Create Slack-native documents

Configuration:

{
  "mcpServers": {
    "slack": {
      "url": "https://mcp.slack.com/mcp",
      "transport": "sse"
    }
  }
}

OAuth flow handles authentication. First connection opens browser for workspace authorization.

Use Cases:

  • Automated standup summaries
  • Cross-channel search for context
  • Posting deployment notifications
  • Creating documentation from threads

Limitations: Rate limits apply. Bot must be invited to private channels for access.

2. Gmail MCP Server

Maintainer: Google (via claude.ai integration) / Community options
Transport: Remote SSE or local with OAuth

Gmail integration varies by implementation. The claude.ai hosted version uses remote OAuth. Self-hosted options require Google Cloud credentials.

Key Tools:

  • send_email - Compose and send messages
  • search_emails - Gmail query syntax search
  • get_unread_emails - Fetch unread inbox
  • read_email - Get full message content
  • trash_email - Move to trash
  • modify_email - Change labels, mark read/unread

Self-Hosted Configuration:

{
  "mcpServers": {
    "gmail": {
      "command": "npx",
      "args": ["-y", "mcp-google-workspace"],
      "env": {
        "GOOGLE_CLIENT_ID": "your-client-id",
        "GOOGLE_CLIENT_SECRET": "your-secret"
      }
    }
  }
}

Requires OAuth 2.0 credentials from Google Cloud Console. Enable Gmail API in your project.

Use Cases:

  • Drafting responses based on conversation context
  • Finding specific emails for task context
  • Automated email triage and labeling

3. Google Calendar MCP Server

Maintainer: Community (multiple implementations)
Transport: Local stdio with OAuth

Multiple Calendar servers exist. The @cocal/google-calendar-mcp package offers multi-account support and conflict detection.

Key Tools:

  • list_events - Upcoming events with date filtering
  • create_event - Schedule with attendees
  • update_event - Modify existing events
  • delete_event - Remove events
  • get_freebusy - Check availability
  • search_events - Find by title/description

Configuration:

{
  "mcpServers": {
    "google-calendar": {
      "command": "npx",
      "args": ["@cocal/google-calendar-mcp"],
      "env": {
        "GOOGLE_OAUTH_CREDENTIALS": "/path/to/credentials.json"
      }
    }
  }
}

Use Cases:

  • Scheduling meetings from natural language
  • Pre-meeting briefing (combine with Gmail search)
  • Availability checking across calendars
  • Automated calendar cleanup

4. Notion MCP Server

Maintainer: Notion (official)
Transport: Remote with OAuth or local with API key

Notion's official MCP server optimizes for token efficiency with Markdown-based responses. The hosted version at developers.notion.com handles OAuth automatically.

Key Tools:

  • create_page - New pages with content
  • search_content - Search across workspace
  • update_database_item - Modify database entries
  • retrieve_page - Get page content as Markdown
  • query_database - Filter and sort database views
  • append_blocks - Add content to existing pages

Local Configuration:

{
  "mcpServers": {
    "notion": {
      "command": "npx",
      "args": ["-y", "@notionhq/notion-mcp-server"],
      "env": {
        "NOTION_API_TOKEN": "secret_..."
      }
    }
  }
}

Create an internal integration at notion.so/my-integrations. Connect pages to the integration for access.

Use Cases:

  • Generating documentation from conversation
  • Project database updates
  • Knowledge base search
  • Meeting notes from transcripts

Limitations: Integration must be explicitly connected to each page/database.

5. Atlassian MCP Server (Jira + Confluence)

Maintainer: Atlassian (official remote beta)
Transport: Remote SSE with OAuth

Atlassian's remote MCP server connects Jira and Confluence to AI tools. Built on Cloudflare's Agents SDK with enterprise-grade OAuth.

Key Tools:

  • Jira: create issues, search with JQL, update status, add comments
  • Confluence: search pages, create content, update documentation

Configuration:

{
  "mcpServers": {
    "atlassian": {
      "url": "https://mcp.atlassian.com/v1/sse",
      "transport": "sse"
    }
  }
}

OAuth connects your Atlassian Cloud instance.

Community Alternative (self-hosted):

{
  "mcpServers": {
    "jira": {
      "command": "npx",
      "args": ["-y", "@mcp-devtools/jira"],
      "env": {
        "JIRA_URL": "https://yourworkspace.atlassian.net",
        "JIRA_API_MAIL": "[email protected]",
        "JIRA_API_KEY": "your-api-token"
      }
    }
  }
}

Use Cases:

  • Creating tickets from bug reports
  • Searching for related issues
  • Updating ticket status from code commits
  • Generating release notes from completed tickets

Development & DevOps

6. GitHub MCP Server

Maintainer: GitHub
Transport: Local stdio

GitHub's MCP server is the gold standard for code-aware agents. Full repository access with identity-based permissions.

Key Tools:

  • create_pull_request - Open PRs with diff
  • get_issue - Fetch issue details
  • list_files - Repository contents
  • search_code - Code search across repos
  • create_issue - File new issues
  • get_workflow_runs - CI/CD status

Configuration:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    }
  }
}

Use Cases:

  • Code review assistance
  • Issue triage and assignment
  • PR description generation
  • Automated documentation updates

7. Sentry MCP Server

Maintainer: Sentry (official)
Transport: Remote SSE with OAuth

Sentry's MCP server puts error context directly in your development workflow. Remote-hosted with no local installation. Essential for teams running production AI evaluations.

Key Tools:

  • Error and issue querying
  • Project and organization access
  • Seer analysis integration (AI debugging)
  • Performance telemetry

Configuration:

{
  "mcpServers": {
    "sentry": {
      "url": "https://mcp.sentry.dev/mcp",
      "transport": "sse"
    }
  }
}

OAuth connects to your Sentry organization. Select tool groups during auth.

Use Cases:

  • Debugging production errors with full context
  • Identifying error patterns
  • Connecting code changes to error resolution
  • Feeding error data into evaluation workflows

8. Linear MCP Server

Maintainer: Community (@mcp-devtools/linear)
Transport: Local stdio

Linear integration for fast-moving development teams.

Key Tools:

  • Issue creation and updates
  • Project and cycle management
  • Search and filtering
  • Status transitions

Configuration:

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "@mcp-devtools/linear"],
      "env": {
        "LINEAR_API_KEY": "lin_api_..."
      }
    }
  }
}

Generate API key in Linear: Settings > API > Create Key.

Use Cases:

  • Sprint planning assistance
  • Issue creation from conversations
  • Status updates from development context

Databases

9. PostgreSQL MCP Server

Maintainer: Google (Toolbox) / Community options
Transport: Local stdio

Multiple PostgreSQL servers exist. Google's MCP Toolbox for Databases provides enterprise-grade access with query safety controls.

Toolbox Configuration:

{
  "mcpServers": {
    "postgres": {
      "command": "./toolbox",
      "args": ["--prebuilt", "postgres", "--stdio"],
      "env": {
        "POSTGRES_HOST": "localhost",
        "POSTGRES_PORT": "5432",
        "POSTGRES_DATABASE": "mydb",
        "POSTGRES_USER": "user",
        "POSTGRES_PASSWORD": "password"
      }
    }
  }
}

Postgres MCP Pro (crystaldba):

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@crystaldba/postgres-mcp"],
      "env": {
        "DATABASE_URI": "postgresql://user:pass@host:5432/db"
      }
    }
  }
}

Key Tools:

  • execute_sql - Run queries with safety controls
  • list_schemas - Database structure
  • describe_table - Column info and constraints
  • explain_query - Query plan analysis

Use Cases:

  • Natural language database queries
  • Schema exploration
  • Query optimization suggestions
  • Data analysis workflows
  • Generating training data for model fine-tuning

Security Note: Use read-only credentials for production databases. Never expose write access to untrusted agents.

10. MySQL MCP Server

Maintainer: Google (Toolbox) / Community
Transport: Local stdio

Configuration:

{
  "mcpServers": {
    "mysql": {
      "command": "./toolbox",
      "args": ["--prebuilt", "mysql", "--stdio"],
      "env": {
        "MYSQL_HOST": "localhost",
        "MYSQL_DATABASE": "mydb",
        "MYSQL_USER": "user",
        "MYSQL_PASSWORD": "password"
      }
    }
  }
}

Same tools as PostgreSQL. Supports both MySQL 8.x and MariaDB.

11. MongoDB MCP Server

Maintainer: MongoDB (official)
Transport: Local stdio

MongoDB's official server with full Atlas support.

Key Tools:

  • Collection operations
  • Aggregation pipelines
  • Index management
  • Query building

Configuration:

{
  "mcpServers": {
    "mongodb": {
      "command": "npx",
      "args": ["-y", "@mongodb/mcp-server"],
      "env": {
        "MONGODB_URI": "mongodb+srv://..."
      }
    }
  }
}

12. SQLite MCP Server

Maintainer: Anthropic (reference)
Transport: Local stdio

The reference SQLite server ships with the official MCP repository. Useful for local development and small-scale data.

Configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sqlite", "/path/to/database.db"]
    }
  }
}

13. Supabase MCP Server

Maintainer: Community
Transport: Local stdio

Combines PostgreSQL access with Supabase-specific features like auth and storage.

Key Tools:

  • Database queries via PostgREST
  • Storage bucket management
  • User authentication
  • Realtime subscriptions

Configuration:

{
  "mcpServers": {
    "supabase": {
      "command": "npx",
      "args": ["-y", "supabase-mcp-server"],
      "env": {
        "SUPABASE_URL": "https://xxx.supabase.co",
        "SUPABASE_SERVICE_KEY": "eyJ..."
      }
    }
  }
}

14. DBHub (Multi-Database)

Maintainer: Bytebase
Transport: Local stdio

Zero-dependency server supporting PostgreSQL, MySQL, SQL Server, MariaDB, and SQLite through a single interface.

Configuration:

{
  "mcpServers": {
    "dbhub": {
      "command": "npx",
      "args": ["-y", "dbhub", "--dsn", "postgresql://user:pass@host:5432/db"]
    }
  }
}

Token-efficient design with just two MCP tools. Good for context-limited scenarios and teams focused on LLM cost optimization.

Browser Automation

15. Playwright MCP Server

Maintainer: Microsoft (official)
Transport: Local stdio

Microsoft's official Playwright MCP brings browser automation to AI agents. Uses accessibility snapshots for efficient interaction without screenshots.

Key Tools:

  • browser_navigate - Go to URL
  • browser_click - Click elements
  • browser_fill - Form input
  • browser_screenshot - Capture page
  • browser_evaluate - Run JavaScript
  • browser_snapshot - Accessibility tree

Configuration:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp"]
    }
  }
}

Browser installs automatically on first use. Supports Chromium, Firefox, WebKit.

Environment Variables:

  • PLAYWRIGHT_MCP_BROWSER - chrome, firefox, webkit, msedge
  • PLAYWRIGHT_MCP_HEADLESS - true/false
  • PLAYWRIGHT_MCP_CAPS - vision, pdf, devtools

Use Cases:

  • E2E testing from natural language
  • Web scraping and data extraction
  • Form automation
  • Visual regression testing

16. Puppeteer MCP Server

Maintainer: Anthropic (reference)
Transport: Local stdio or Docker

Puppeteer-based browser automation. Simpler than Playwright but limited to Chromium.

Configuration:

{
  "mcpServers": {
    "puppeteer": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
    }
  }
}

Docker Deployment:

{
  "mcpServers": {
    "puppeteer": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "--init", "-e", "DOCKER_CONTAINER=true", "puppeteer-mcp"]
    }
  }
}

Supports Cloudflare Workers deployment for cloud-based automation.

Vector Databases & Memory

17. Qdrant MCP Server

Maintainer: Qdrant
Transport: Local stdio

High-performance vector search for agent memory and RAG applications.

Key Tools:

  • Collection management
  • Vector upsert and search
  • Filtering and payload queries
  • Batch operations

Configuration:

{
  "mcpServers": {
    "qdrant": {
      "command": "npx",
      "args": ["-y", "qdrant-mcp"],
      "env": {
        "QDRANT_URL": "http://localhost:6333",
        "QDRANT_API_KEY": "optional-key"
      }
    }
  }
}

Use Cases:

  • Semantic search in RAG pipelines
  • Agent long-term memory
  • Document similarity matching
  • Context retrieval for open source models

For production RAG implementations, vector databases provide the retrieval layer.

18. Chroma MCP Server

Maintainer: Community
Transport: Local stdio

Open-source embedding database optimized for AI applications.

Key Tools:

  • Collection CRUD
  • Document add/query/delete
  • Metadata filtering
  • Full-text search

Configuration:

{
  "mcpServers": {
    "chroma": {
      "command": "npx",
      "args": ["-y", "chroma-mcp"],
      "env": {
        "CHROMA_HOST": "localhost",
        "CHROMA_PORT": "8000"
      }
    }
  }
}

Supports both ephemeral (in-memory) and persistent storage modes.

19. Memory MCP Server

Maintainer: Anthropic (reference)
Transport: Local stdio

Knowledge graph-based persistent memory. Stores entities and relationships, not raw vectors.

Key Tools:

  • create_entities - Add nodes
  • create_relations - Connect nodes
  • search_nodes - Find by name/type
  • open_nodes - Get connected context

Configuration:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}

Good for tracking project context, people, and relationships across conversations.

Cloud & Infrastructure

20. AWS MCP Servers

Maintainer: AWS / Community
Transport: Local stdio

Multiple AWS MCP servers exist for different services.

AWS Documentation Server:

{
  "mcpServers": {
    "aws-docs": {
      "command": "npx",
      "args": ["-y", "aws-mcp-server"]
    }
  }
}

Exposes AWS documentation, billing data, and service metadata.

AWS Bedrock:

{
  "mcpServers": {
    "bedrock": {
      "command": "npx",
      "args": ["-y", "@aws/bedrock-mcp"],
      "env": {
        "AWS_ACCESS_KEY_ID": "...",
        "AWS_SECRET_ACCESS_KEY": "...",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

For teams deploying enterprise AI on AWS, these servers provide infrastructure integration.

21. Cloudflare MCP Server

Maintainer: Cloudflare
Transport: Edge (Workers)

Cloudflare's edge-native MCP distributes agent computation globally. Lower latency for geographically distributed users.

Key Features:

  • Workers KV access
  • D1 database queries
  • R2 storage operations
  • Edge-native deployment

Use Cases:

  • Globally distributed agents
  • Low-latency tool calls
  • Edge caching for frequent operations

22. Google Cloud MCP Servers

Maintainer: Google
Transport: Remote (managed)

Google's managed MCP servers cover AlloyDB, Cloud SQL, Spanner, Firestore, and Bigtable.

Cloud SQL Configuration:

{
  "mcpServers": {
    "cloudsql": {
      "url": "https://mcp.googleapis.com/cloud-sql/v1",
      "transport": "sse"
    }
  }
}

No infrastructure management. Configure endpoint and authenticate.

File & Storage

23. Filesystem MCP Server

Maintainer: Anthropic (reference)
Transport: Local stdio

Secure file operations with configurable access controls.

Key Tools:

  • read_file - Get file contents
  • write_file - Create/update files
  • list_directory - Browse folders
  • search_files - Pattern matching
  • get_file_info - Metadata

Configuration:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/allowed/path"]
    }
  }
}

Access restricted to specified paths. Multiple paths can be allowed.

24. Google Drive MCP Server

Maintainer: Community
Transport: Local stdio with OAuth

Access Drive files, folders, and shared drives.

Key Tools:

  • Search files
  • Read document contents
  • Upload files
  • Folder navigation
  • Permission management

Configuration:

{
  "mcpServers": {
    "gdrive": {
      "command": "npx",
      "args": ["-y", "google-drive-mcp"],
      "env": {
        "GOOGLE_CLIENT_ID": "...",
        "GOOGLE_CLIENT_SECRET": "..."
      }
    }
  }
}

Requires OAuth credentials from Google Cloud Console with Drive API enabled.

Search & Web

25. Brave Search MCP Server

Maintainer: Brave (official)
Transport: Local stdio

Web and local search through Brave's privacy-focused search API.

Key Tools:

  • web_search - General web search
  • local_search - Location-based results

Configuration:

{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@anthropic/brave-search-mcp"],
      "env": {
        "BRAVE_API_KEY": "..."
      }
    }
  }
}

Get API key from api.search.brave.com.

Use Cases:

  • Real-time information retrieval
  • Research augmentation
  • Fact-checking

Authentication Patterns

MCP servers use three main auth patterns:

API Keys: Simple, set via environment variable. Used by GitHub, Linear, databases.

OAuth 2.0: Browser-based flow for user consent. Used by Slack, Notion, Google services. Store refresh tokens securely.

No Auth: Local-only servers with implicit trust. Filesystem, SQLite, Puppeteer.

For enterprise deployments with data security requirements, audit which servers send data externally.

Security Considerations

MCP servers execute arbitrary actions. Treat them with appropriate caution:

  1. Principle of Least Privilege - Use read-only credentials when possible
  2. Path Restrictions - Limit filesystem access to necessary directories
  3. Network Isolation - Run database servers in private networks
  4. Audit Logging - Log MCP tool invocations for compliance and observability
  5. Token Rotation - Rotate API keys and OAuth tokens regularly

Production deployments should implement policy enforcement. Tools like Pomerium add zero-trust controls to MCP servers.

Running Multiple Servers

Most workflows require multiple servers. Example multi-server configuration:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "..." }
    },
    "slack": {
      "url": "https://mcp.slack.com/mcp",
      "transport": "sse"
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@crystaldba/postgres-mcp"],
      "env": { "DATABASE_URI": "..." }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/projects"]
    }
  }
}

Each server loads independently. Clients see combined tool sets.

Debugging MCP Servers

Common issues and fixes:

Server won't start:

  • Check command path exists
  • Verify environment variables are set
  • Run command manually to see errors

Authentication fails:

  • Confirm API key/token is valid
  • Check OAuth scopes match required permissions
  • Verify token hasn't expired

Tools not appearing:

  • Restart client after config changes
  • Check server logs for initialization errors
  • Verify JSON syntax in config file

Slow responses:

  • Network latency for remote servers
  • Database query performance
  • Rate limiting on external APIs

Building Custom MCP Servers

When existing servers don't cover your use case, build custom ones using official SDKs:

Python (FastMCP):

from fastmcp import FastMCP

mcp = FastMCP("my-server")

@mcp.tool()
def my_custom_tool(param: str) -> str:
    """Tool description for the AI."""
    return f"Processed: {param}"

mcp.run()

TypeScript:

import { Server } from "@modelcontextprotocol/sdk/server/index.js";

const server = new Server({ name: "my-server", version: "1.0.0" });

server.setRequestHandler(ListToolsRequestSchema, async () => ({
  tools: [{ name: "my_tool", description: "...", inputSchema: {...} }]
}));

For complex integrations, the agentic AI patterns guide covers multi-tool orchestration.

Choosing Servers by Use Case

Developer workflows: GitHub + Sentry + Linear + Filesystem
Knowledge work: Notion + Slack + Gmail + Google Drive
Data analysis: PostgreSQL + Brave Search + Filesystem
Testing automation: Playwright + GitHub + Slack
Customer support: Slack + Notion + Database

Start with two or three servers. Add more as workflows mature.

For teams building custom AI applications, platforms like Prem Studio provide integrated development workflows that complement MCP server capabilities.

FAQ

How many MCP servers can I run simultaneously?

No hard limit, but each server consumes context. Running 10+ servers may exhaust token budgets in longer conversations. Teams using small language models face tighter context limits. Enable only servers needed for current tasks.

Do MCP servers work with OpenAI and other providers?

Yes. Since OpenAI adopted MCP in March 2025, most major AI tools support the protocol. Configuration syntax varies by client.

Can I self-host official MCP servers?

Depends on the server. Anthropic's reference servers are MIT-licensed. Notion, Slack, and Atlassian official servers are remote-only. Community alternatives often support self-hosting.

How do I handle rate limits?

Implement exponential backoff in your agent logic. Some servers (GitHub, Slack) return rate limit headers. Monitor usage and add delays between rapid operations.

Are MCP connections secure?

Local stdio servers communicate over process pipes (secure). Remote SSE/HTTP servers use TLS. OAuth tokens should be stored securely, not in plain-text configs.

What's the difference between MCP and function calling?

Function calling requires vendor-specific integration per tool. MCP provides a universal standard, so one server works with any MCP-compatible client. MCP also supports resources and prompts beyond just tools.

Subscribe to Prem AI

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
[email protected]
Subscribe