AI & Technology
How AI Agents Connect to Your Tools (MCP, Explained)
MCP is an open standard for connecting AI agents to tools. What it is, which of your tools your agents can reach, and how a link is built inside a plan.
AI agents connect to tools through an open standard called MCP, not through proprietary connectors that tie you to one platform. If a tool publishes an MCP server, your agents can reach it; if it has an API or a dashboard, we build the link. Either way it is part of the Business Partnership plan. This page explains what MCP is and how that works.
What Is MCP?
MCP stands for Model Context Protocol. Anthropic created it as an open standard that defines how AI agents communicate with external tools and data sources. Think of it as a shared language that lets an agent talk to any system that speaks the same protocol.
Before MCP, connecting an AI agent to an external tool meant writing proprietary glue code: custom API bridges specific to one vendor's implementation. If you switched tools or the vendor changed their API, the connection broke and someone had to rebuild it. MCP replaces that with a shared standard. Any tool that publishes an MCP server can be reached by any MCP-compatible agent — the same way any website can be visited by any standards-compliant browser.
Vercel ships a remote MCP server that shows what this looks like: OAuth-authenticated tool access over an open protocol. See the technical detail at vercel.com/docs/mcp and the protocol explainer at vercel.com/blog/model-context-protocol-mcp-explained.
Why MCP Matters for Ownership
An open protocol provides a shared way to connect agents and tools. It does not guarantee compatibility, independent hosting or portability of every connection. Access, credentials, third-party systems and CNX platform dependencies still need to be assessed for the agreed work and any departure handoff.
Open standards accumulate value for the owner of the infrastructure, not for the platform that built the connectors.
What Your Agents Can Reach
MCP is how agents reach tool servers that publish one. Tools such as QuickBooks, HubSpot, Microsoft 365, Square and Stripe publish MCP servers or are adding them. We manage that registry for you, and tool connections are set up per engagement.*
Running now for clients:
- Email — send and receive through your own mailbox
- Google Chat — message your AI team where you already work
- SMS and WhatsApp — through Twilio
- Files — documents you share with your AI team
- Stripe — for billing only
Our ad-account work reads data only. There is no tool we rule out before looking at it.
Connecting Your Own Tools
If it has an API, an MCP server or a dashboard, we can connect it.* That covers accounting, CRM, booking and POS tools such as QuickBooks, HubSpot, Square and Toast. Connecting your tools is part of the Business Partnership plan. If a tool offers no way to connect, we say so up front.
* Some tools without native integration capabilities may require longer timelines and workarounds such as computer use, where an agent works in the tool's own screen under a person's supervision. We confirm the scope in writing before we build, and a few tools give us no way in. Tool names belong to their owners; naming one means we can work with it, not that they endorse or certify us. See what we connect.
Ownership of qualifying customer-specific custom IP starts in subscription month 13. Whether a custom connector is transferable, and which dependencies it needs, must be specified. CNX's platform and underlying technology remain CNX's.
What This Means for Your Security
Every MCP call your agents make is governed by scoped permissions, role-based access controls, and a reviewable audit trail. An agent that is allowed to read from one connection cannot use it to write somewhere else. Permissions are defined per agent, per tool, per action — not granted broadly.
Tool calls are logged so you can review what your agents accessed and when.
Frequently Asked Questions
Is MCP a standardized protocol, or something Connected Spaces invented?
MCP is an open standard created by Anthropic. It is publicly documented and implemented by multiple vendors. Connected Spaces uses it because it is the right architecture for owner-controlled infrastructure — not because it gives us proprietary lock-in.
Can I see what tools my agents have access to?
Yes. We set up each agent with a defined list of connections and the permissions granted for each, and you can ask us for that list at any time.
Can I revoke an agent's access to a specific tool?
Yes. Connections are configured per agent and can be removed or restricted without affecting other agents or connections.
Does my data go to the AI company?
MCP governs the communication between your agents and your tools. Whatever the model does see is covered by the business terms we hold with every provider: your data is not used to train their models. Stricter data-residency or isolation options are available where regulation requires.
What if a tool I use doesn't have an MCP server yet?
Tell us which tool. If it has an API or a dashboard, the link is part of the Business Partnership plan. We don't promise compatibility before scoping.
MCP, Ownership, and Growing Your Tool Stack
Using an open standard does not by itself make a connection independent of its platform or third-party tools. For a departure handoff, confirm the transferable custom work, credentials, hosting requirements and dependencies. Ownership of qualifying customer-specific custom IP starts in subscription month 13; it does not begin at build completion.
New tool connections are part of the Business Partnership plan, with the scope confirmed in writing before we build them. A published MCP server does not guarantee compatibility with your setup. Custom connector ownership and handoff follow the same qualified month-13 scope above; do not assume every server or dependency transfers.
On data privacy: MCP governs communication between your agents and your tools. Anything the model layer does see is governed by a deployment model we configure to your regulatory needs — your data is not used to train the providers' models under the business terms we hold, and stricter data-residency or isolation models are available where regulation requires.
See which plans include your AI team
Go Deeper
Related
- Message your AI team over WhatsApp, Google Chat, SMS or email — No new app, no new login — direct your AI Agents from channels you already use.
- From chat to execution — How one message becomes a completed, traceable business task.
- How your AI team coordinates — @-mentions, agent-to-agent handoffs, and the HITL approval boundary.
- AI Agents vs. rented infrastructure — Why AI agents need a custom data foundation to be truly useful.