AI assistants are good at reasoning and writing, but on their own they cannot see your systems. Before MCP, every connection between an AI tool and a piece of software was a custom integration. MCP gives them a common language, so one server can work with many AI clients and one client can use many servers.
How MCP works
There are three parts:
| Part | Role | Example |
|---|---|---|
| MCP host or client | The AI application the person uses | Claude, Claude Code, or another MCP-capable assistant |
| MCP server | A program that exposes tools and data | A CRM’s MCP server |
| Tools | Named actions the server offers, with inputs | ”search records”, “update a record”, “add a note” |
The client connects to the server, asks what tools it offers, and then calls them when the conversation needs them. The server does the work using its own permissions and returns the result to the AI.
Why it matters for business software
- Your assistant can act where your data lives. Ask your AI tool to find the newest unclaimed leads and claim them, and it does it in the CRM rather than describing how.
- No bespoke integration per AI vendor. A software company builds one MCP server and it works with any compatible client.
- Agents become practical. Multi-step jobs, such as working through a queue and leaving notes, become something an AI agent can do end to end.
Safety considerations
Connecting an AI to real systems deserves the same care as giving a new colleague access:
- Scope the credentials. Use a key that can only read when reading is all the agent needs.
- Match human permissions. An agent should never be able to do more than the person or role it acts for.
- Keep revocation easy. If a key leaks or an agent misbehaves, you should be able to cut it off immediately.
- Review what agents change, at least until you trust the workflow.
Common misconceptions
- MCP is not a model. It is the plumbing between a model and your tools.
- MCP does not bypass security. The server decides what is allowed, based on the credentials it is given.
- It is not limited to one vendor. It is an open specification with clients and servers from many companies.
How GuardenerAI handles it
- GuardenerAI runs an MCP server over streamable HTTP, so Claude and other MCP-capable agents can work with your CRM. Setup is on the Developer API and MCP page.
- Its tools: search records, get a record, create a record, update a record, add a note, list a worklist, claim a record and release a record.
- Agents connect with the same API keys as the Developer API. Keys are read-only or read and write, can never do more than a Member of the organisation, and stop working the moment they are revoked.
- Inside the app, the built-in AI assistant works on the same records without any setup.