MCPIntegrationGuide:ConnectYourToolstoStrawberry

MCP is the protocol everyone in AI started talking about in 2025, and most explanations stop at "it's like USB-C for AI." That metaphor is fine, but it does not tell you what to actually do with it, or why connecting a tool to an AI client changes anything about your work.

This guide answers the practical questions: what MCP (Model Context Protocol) is, how servers and clients fit together, what you get when you connect tools to an AI client, and how Strawberry uses MCP plus connected apps plus your real browser tabs so an agent can finish a workflow instead of just answering a question about it. If you came here to decide whether MCP alone is enough, or whether you also need a place for the work to happen, that is the honest comparison below.

The short answer

MCP is a genuinely good open standard. It gives any AI application a uniform way to call tools, read data, and reuse prompts from an external system, so you stop hand-wiring a custom integration for every app. But MCP is a connection layer, not a worker. It defines how an agent can reach your CRM or your repo; it does not prep the call, log the record, or run the workflow every morning. Strawberry uses MCP and your connected apps as inputs, then does the work across the browser sessions you are already logged into. If you want the protocol, MCP is the answer. If you want the job finished, you need MCP plus a place the agent can act.

Where a raw MCP setup wins

Before the Strawberry pitch, here is where wiring MCP servers directly into a desktop client or your own code is genuinely the better choice.

Full control and portability

MCP is an open standard, so a server you build or install works with any MCP-compatible client, not just one product. If you care about avoiding lock-in, or you want the same CRM server to serve a local script today and a different agent next year, a raw setup keeps that door open. You own the config, the auth, and the tool definitions.

Deep, custom servers for your own systems

If your most valuable context lives in an internal database or a service with no public app, writing your own MCP server is the clean way to expose exactly the tools you want, with the output shapes you want. A purpose-built server you control will usually beat a generic connector for that one system.

Developer workflows and local tooling

For engineers, MCP servers running next to a coding assistant - file access, a test runner, a database query tool - are fast, scriptable, and live where the work already is. If your whole workflow is in a terminal and an editor, you may not need a browser in the loop at all.

Where Strawberry wins

Strawberry is a browser with built-in AI, so MCP is one of several ways its agents reach your work, not the only one. That mix is what turns a tool call into a finished task.

Connected apps and MCP, together, with no plumbing

Strawberry already ships native connections to the apps most people use - Gmail, Calendar, Sheets, Slack, Notion, HubSpot, Salesforce, GitHub and more - and you can register any MCP server on top. So you are not choosing between "easy connectors" and "the MCP ecosystem." You connect the apps you have, add the servers you need, and the agent uses whichever fits the step. Connecting one is a guided action in the app, not a JSON file you edit by hand.

It acts in the sessions you are already logged into

This is the part a pure MCP client cannot do. A lot of real work lives in web apps, admin panels, and dashboards that have no clean API and no MCP server. Strawberry runs in your browser, logged into those sessions, so when MCP and APIs run out the agent can still open the tab and do the thing - pull the number off the dashboard, fill the form, click through the panel - and stitch it back into the workflow.

Companions, Skills, and Routines turn a connection into a habit

A connected tool only pays off if something uses it repeatedly. A Companion is a persistent teammate that remembers your context, a Skill is a workflow you save once, and a Routine runs that Skill on a schedule. So "connect my CRM and calendar" becomes "every weekday at 7am, prep briefs for today's calls, find deals gone quiet, and draft the next touch" - using your connected apps and MCP servers, with no re-asking. With Team share, one bill covers shared Companions and Skills across the team.

How it works in practice

The setup is deliberately short. Open Strawberry, connect the apps you already use, and add any MCP server you need by registering its address - the app walks you through auth, so there is no config file to maintain. From there you describe the job in plain language and let the agent combine connected-app calls, MCP tools, browser tabs, and your files to finish it.

A few concrete shapes this takes:

• A sales operator connects a CRM, Gmail, and calendar, adds an MCP lead-data server, then asks Strawberry to build a shortlist, research each account, draft outreach, and log every follow-up.
• A recruiter connects an ATS and calendar, adds an MCP sourcing server, then has Strawberry prepare interview briefs and draft personalized candidate messages.
• A founder connects analytics, email, and docs, then turns a daily operating brief into a Routine that runs before they open their laptop.

On permissions: connections are scoped and you stay on the send button by default. Strawberry drafts the email, the CRM update, or the message for you to review rather than firing it off on its own. Treat MCP servers the way you treat any integration - connect ones you trust, give them the access the job needs and no more, and keep destructive actions behind your own confirmation.

Keep going

• what is an agentic browser - why a browser with built-in AI can act across your real tabs, not just answer questions.
• operations use cases - concrete workflows where connected tools plus MCP turn into finished operational work.
• Strawberry vs Zapier - how agent-driven work in the browser differs from fixed automation pipelines.