MCP servers let Claude reach beyond your folder and into Notion, Gmail, Slack, Stripe, and more. One connection each. No copy-pasting between apps. This is what "connected AI" actually means.
What this episode covers
Without connectors, Claude can only work with what's in front of it. With them, it can pull from Notion, check your calendar, post to Slack, query Stripe, all in one workflow, without you touching another tab.
A universal adapter. The USB port that connects Claude to everything else.
The Model Context Protocol (MCP) is an open standard Anthropic published in November 2024. It defines a common language for AI assistants to connect to external tools. Think of it like USB: before USB, every device needed its own proprietary port. After USB, one standard connected everything. MCP does the same thing for AI: any tool that implements it can speak to any AI assistant that supports it.
An MCP server is the connector piece: a small program that sits between Claude and a specific service (Notion, Slack, Stripe, GitHub…) and gives Claude a set of structured actions it can call. When you install a Notion MCP server, Claude gains the ability to search pages, read databases, and create entries, all from inside a conversation, without you manually extracting and pasting.
The MCP client in Claude Code manages all the connections. When Claude needs information from Notion, it sends a structured request to the Notion MCP server, which handles authentication and returns the data. Claude never stores your credentials. That stays inside the server.
MCP adoption moved faster than almost any protocol in recent memory. OpenAI adopted it in March 2025. Google confirmed Gemini support in April 2025. As of 2026 there are thousands of MCP servers covering virtually every major SaaS tool. If a service exists, there's probably an MCP server for it, and if there isn't, there will be soon.
The practical consequence: the workflow of "paste from app A into Claude, paste result into app B" is going away. Operators who connect their tools now are running workflows in 2026 that their competitors are still doing manually.
One JSON file that tells Claude which servers to load. You rarely need to edit it by hand.
When Claude Code starts a session, it looks for a file called .mcp.json at the root of your project folder. This file is a list of MCP servers to load: their names, how to connect to them, and any credentials they need. It's the registry of "which tools does Claude have access to right now?"
{
"mcpServers": {
"notion": {
"type": "http",
"url": "https://mcp.notion.com/mcp"
},
"google-workspace": {
"type": "http",
"url": "https://mcp.googleapis.com/mcp"
},
"stripe": {
"type": "http",
"url": "https://mcp.stripe.com"
}
}
}
Each entry has a name you pick (like "notion") and a connection type. HTTP servers (the common case) just need a URL: Claude connects over the internet, authenticates via OAuth, and can start using the service. No local process to run, no installation beyond this file.
| Scope | File location | Who can see it | When to use |
|---|---|---|---|
| Project | .mcp.json in your project root |
Anyone with access to the folder | Shared team tools: Notion, Slack, GitHub for this project |
| User | ~/.claude.json (your home folder) |
Only you, across all projects | Personal tools: your Gmail, your calendar |
| Local | User-level but scoped to one project | Only you, this project only | Project-specific credentials you don't want to share |
.mcp.json and placed at the root of the folder you opened in Claude Code, not inside .claude/. If your servers aren't loading, check the file is at the right level.
In practice, you rarely need to edit this file manually. Most installs today use the CLI command claude mcp add or the app's Connectors panel, which write the file for you. But knowing it exists means you can inspect it, share it with your team, and understand exactly what Claude has access to in any given workspace.
From easiest (double-click) to most flexible (the terminal command). All legitimate paths.
Claude Desktop supports .mcpb files: packaged extensions that install with a double-click. Open Claude Desktop → Settings → Extensions → Browse extensions to find and install official integrations. No JSON editing, no terminal commands. This is the right starting point for non-technical users.
.mcpb packages. Coverage is growing but not every server has one yet. Check here first, fall back to Method 02 if you don't find what you need.
claude mcp add commandFor most modern MCP servers (Notion, Slack, Stripe, Airtable, GitHub) the recommended install is a single terminal command. It writes the .mcp.json entry for you and handles authentication via OAuth.
# General pattern claude mcp add --transport http [name] [server-url] # Examples claude mcp add --transport http notion https://mcp.notion.com/mcp claude mcp add --transport http slack https://mcp.slack.com/mcp claude mcp add --transport http stripe https://mcp.stripe.com claude mcp add --transport http airtable https://mcp.airtable.com/mcp
After running the command, open Claude Code and type /mcp. A browser window opens asking you to authenticate with the service (same as authorizing any app). Once approved, the server is live.
--transport http flag means "connect over the internet" (as opposed to running a local program). That's the whole command.
By default, claude mcp add installs the server for the current project. Add --scope user to make it available across all your projects, useful for personal tools like your own Gmail or calendar.
# Available across all your projects claude mcp add --scope user --transport http google-workspace https://mcp.googleapis.com/mcp # Only for this project (default) claude mcp add --transport http notion https://mcp.notion.com/mcp
Curated for business owners, marketers, and operators. All are real, installable, and actively maintained as of 2026.
Read and create calendar events, search your Gmail inbox, access Drive documents, all in one server. The single highest-leverage install for operators who live in Google's tools. Claude can check your availability, draft email replies, and pull documents into any workflow.
claude mcp add --scope user --transport http google-workspace https://mcp.googleapis.com/mcp
Search pages, read databases, create and update entries. If your team runs on Notion (CRM, content calendar, wiki, project tracker), Claude can now read and write it directly. After install, share the specific Notion pages with the integration via Notion's sharing settings.
claude mcp add --transport http notion https://mcp.notion.com/mcp
Search message history, read channel threads, look up users, send messages. Claude can search for context from last week's discussion, summarize a channel, or post an update, all without you leaving the workspace. Pairs especially well with Routines from Episode 4.
claude mcp add --transport http slack https://mcp.slack.com/mcp
Query customers, payments, subscriptions, and revenue data. Search Stripe's documentation. Claude can pull MRR, check a specific customer's history, or look up a failed payment, all in natural language. Useful for anyone running a subscription business.
claude mcp add --transport http stripe https://mcp.stripe.com
Browse repositories, read issues and pull requests, create issues, search code. Even non-technical users benefit here: Claude can find the issue that was opened about a customer bug, read what the engineering team wrote, and draft a customer response with accurate technical context.
claude mcp add --transport http github https://api.githubcopilot.com/mcp
Note: GitHub requires a Personal Access Token (PAT) passed as a header. The install wizard prompts you through this.
Read and write records, inspect schemas, query tables. If you use Airtable as a CRM, content tracker, or operations database, Claude can now query it directly and write results back. Combine with Routines for automated weekly database health checks.
claude mcp add --transport http airtable https://mcp.airtable.com/mcp
Access feature flags, analyze funnels, query A/B test results, explore user behavior data. Marketers and product managers can ask Claude to analyze PostHog data without writing SQL. Installs via PostHog's built-in wizard that configures itself inside Claude Code.
claude mcp add --transport http posthog https://mcp.posthog.com/mcp
Find, create, update issues, projects, and cycles. Useful for operators who collaborate with engineering teams. Claude can check the status of an issue, create a bug report from a customer complaint, or look up what's in the current sprint, all from a conversation.
claude mcp add --transport http linear https://mcp.linear.app/mcp
Read contacts, deals, companies, and activities. Useful for anyone running sales or marketing pipelines in HubSpot. Claude can pull a contact's full history before a call, summarize deal stage across your pipeline, or create new contacts from conversation notes.
claude mcp add --transport http hubspot https://mcp.hubspot.com/mcp
If you want finer-grained control over Drive separately from the Workspace bundle (searching specific folders, reading Docs or Sheets, creating files in specific locations), a standalone Drive server gives you more targeted access. Use the Google Workspace bundle first; switch to this only if you need separate scopes.
claude mcp add --transport http gdrive https://mcp.googledrive.com/mcp
Note: URL may vary by provider. The Google Workspace bundle at googleapis.com/mcp is the more reliably-sourced option.
A key is a password that proves you're allowed to use a service. Here's how they work and where they go.
You'll encounter API keys in two situations: setting up MCP servers that require key-based auth (instead of OAuth), and using Claude Code's API directly. Neither requires you to understand the underlying technology. Just where to put the key.
echo 'export ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.zshrc source ~/.zshrcPlain English: "add this key to the list of settings my Terminal loads every time it opens." It's then available to Claude Code and any other tool on your machine automatically.
.claude/settings.json in your project:
{
"env": {
"STRIPE_API_KEY": "sk_live_..."
}
}
.env with one key per line:
ANTHROPIC_API_KEY=sk-ant-... NOTION_API_KEY=secret_...Claude Code reads
.env files automatically.
.env files automatically, which means if your .env contains database passwords, AWS keys, or other sensitive credentials alongside API keys, Claude has access to them too. If that concerns you, add a deny rule in .claude/settings.json:
{
"permissions": {
"deny": ["Read(./.env)", "Read(./.env.*)"]
}
}
This blocks Claude from reading the file while still letting you use it for other local tooling.
Most modern MCP servers use OAuth: you click "authorize" in a browser and the server handles authentication. For servers that require an API key instead (typically older or self-hosted ones), the key goes in the env block of that server's entry in .mcp.json:
{
"mcpServers": {
"my-server": {
"command": "npx",
"args": ["-y", "@example/mcp-server"],
"env": {
"MY_SERVICE_API_KEY": "key-goes-here"
}
}
}
}
The key stays inside the MCP server's process and is never sent to Claude directly. Claude sees the results of API calls, not the credentials used to make them.
One account, one API key, 300+ AI models. The aggregator layer worth knowing about.
OpenRouter (openrouter.ai) is a unified API gateway that gives you access to over 300 AI models (Claude, GPT-5.5, Gemini, Llama, Mistral, and hundreds more) through a single endpoint and one API key. You sign up once, put credit on your account, and can switch between models by changing a single line.
Try Llama, Qwen, or Mistral for a specific task without creating separate accounts. Compare outputs. Find the right model for each workflow without vendor lock-in.
Some tasks don't need Claude Opus. Route simpler classification or formatting tasks to cheaper models automatically, keeping Claude's quota for the work that actually needs it.
If one provider is down or rate-limited, OpenRouter routes to the next available one. Less babysitting, fewer interruptions to automated workflows.
Instead of managing separate accounts and invoices for Anthropic, OpenAI, and Google: one OpenRouter account, one monthly bill, pay-as-you-go per token.
OpenRouter is not a replacement for Claude Code's default Anthropic connection. Claude Code is optimized to work directly with Anthropic's API. Routing through OpenRouter adds latency and loses some Claude-specific features. Use OpenRouter when you're building something that needs to work across multiple models, not as a default replacement for your Claude subscription.
Several models on OpenRouter offer free tiers: real free access with rate limits, no credit card required. As of 2026, these include some Llama and Qwen variants. Useful for testing workflows before committing to a paid model.
There's a community MCP server that connects Claude Code to OpenRouter, and OpenRouter has its own Claude Code integration guide. But for most non-technical users, the simplest path is: start with your standard Claude subscription, use OpenRouter only when you have a specific cost-or-model reason. Don't add complexity before you need it.
New practical guides for European SMEs adopting AI, delivered when they drop. No spam, unsubscribe anytime.
Want this done for you, not just explained? Vectimo runs the AI adoption end to end.
Work with Vectimo →