MCP Connectors Explained: Connect Your Apps to Claude Cowork Mode

What Is MCP? (Plain English)

MCP stands for Model Context Protocol. It is an open standard created by Anthropic that allows AI models like Claude to securely communicate with external tools and data sources. Think of it like a universal plug socket for AI: instead of each app needing its own custom integration, MCP provides one consistent way for Claude to connect to any service that supports it.

In practical terms: when you install an MCP connector for Slack, Claude can read your messages and send replies on your behalf. When you install one for Google Drive, Claude can open, read, and create files in your Drive. The connector is a secure bridge between Claude and the app.

MCP Connectors vs Plugins - What’s the Difference?

MCP connectors are individual connections to specific apps (one connector = one integration). Plugins are bundles that can contain multiple connectors, custom skills, and settings packaged for a specific use case. A "Sales plugin" might bundle a CRM connector, an email connector, and an outreach-drafting skill - all in one install.

Which Apps Can I Connect?

The most commonly used MCP connectors include:

• Communication: Slack, Gmail, Microsoft Teams, Outlook
• Productivity: Google Drive, Notion, Obsidian, OneDrive
• Project Management: Asana, Linear, Jira, Trello
• CRM & Sales: HubSpot, Salesforce, Pipedrive
• Data: Google Sheets, Airtable, PostgreSQL, BigQuery
• Dev Tools: GitHub, GitLab, Sentry

How to Install Your First MCP Connector

Step 1: Open Claude desktop app and click the gear icon (Settings).

Step 2: Go to the Connectors or MCP section and browse available connectors.

Step 3: Click Install or Connect on the connector you want.

Step 4: Complete the OAuth authorisation - you sign into the app and grant Claude the permissions it needs. You never give Claude your password directly.

Step 5: The connector appears as active. You can now use it in Cowork mode conversations.

What Can You Do Once Connected?

With Slack connected:

With Gmail connected:

With Notion connected:

Is It Safe?

OAuth authorisation: You never give Claude your password. The connection uses industry-standard OAuth - the same "Sign in with Google" process you use everywhere.

Local processing: Cowork mode runs locally on your machine. Data from connected apps is fetched and processed locally, not sent to a remote server for storage.

You control permissions: Revoke connector access at any time through Claude settings or directly through the connected app’s own security page.

Which Connector Should You Start With?

Start with the app you spend the most time in. For most people that is email (Gmail/Outlook) or files (Google Drive/OneDrive). The payoff is immediate - being able to ask Claude to "summarise the 15 emails I haven’t read yet and flag anything urgent" saves real time from day one.

Real-World Example: An MCP Server I Built

To make this concrete, here is a real MCP server I shipped to production in May 2026 - one that lets Claude (and any other MCP-compatible AI tool) answer streaming questions for Asian markets with live data.

The problem it solves

When you ask Claude something like "where can I watch Pulang Araw in the Philippines?", the model either guesses or refuses. Streaming availability is highly geo-specific and changes weekly - there is no way for a language model to know it accurately on its own. That is exactly the gap MCP fills: instead of trying to bake streaming data into the model, expose it via a server the model can call at question time.

The build

I scaffolded the package, wrote three tools, and published it to npm - all in about 14 hours on a single Saturday. The three tools the server exposes:

• search_titles(q, country, limit) - fuzzy-search the TMDB-backed catalog with country-aware return links. Used when the user’s query is ambiguous.
• where_to_watch(title, country) - the most-used tool. Returns the OTTs streaming a specific title in a specific country, grouped by stream / free-with-ads / rent / buy.
• whats_new_on(provider, country, limit) - recent additions to a specific OTT in a specific country. Powers "what is new on JioHotstar this week" type queries.

How users install it

Anyone with Claude Desktop can drop a tiny config snippet into their claude_desktop_config.json:

"ottasia": {
  "command": "npx",
  "args": ["-y", "@ottasia/mcp-server"]
}

From the user’s perspective, Claude just "knows" about Asian streaming providers now. No retraining. No fine-tuning. Plug in a 30-line config snippet and the answers go from "I’m not sure where this streams in Bangladesh" to "Streaming on Hoichoi and free-with-ads on MX Player."

Distribution and adoption

The package - @ottasia/mcp-server on npm - is also listed in the official MCP Registry, mcp.so, Smithery, Glama, and awesome-mcp-servers. Five days after publish day it had 515 downloads, with steady 10-20 per day organic growth since. The biggest spike (450 downloads) came on launch day from a single Twitter + Reddit announce.