Connecting tools to your agents
Your agents know what's in your Appricity workspace. With MCP servers they can also use tools from your other systems: read a deal's stage from your CRM during a conversation, check an order in your own database, look something up in your help desk, or open a ticket when a meeting is booked. MCP (the Model Context Protocol) is the open standard many products now use to offer tools to AI assistants, so a system that works with Claude Desktop or Claude Code usually works here too.
How do I add a server?
Open Operations → MCP servers and choose Add server.
- Give it a name. Agents see the name next to each of its tools, so they know where the data comes from.
- Paste the server's address. It must start with
https://. - Choose how Appricity signs in to it:
- Sign in (OAuth) (most hosted servers): you're sent to the other system to sign in and approve Appricity, then brought back.
- API key / bearer token: paste the key the other system gave you. It's stored encrypted and never shown again.
- No sign-in: for a server that needs none.
Appricity then asks the server which tools it offers and lists them. Every tool starts turned off.
Only workspace admins can add servers and choose their tools.
How do I choose which tools are allowed?
Each tool shows what it does, what it takes and whether it can change data. A tool counts as read-only only if the server says so; if it doesn't, Appricity treats it as one that can change data. If a server says a tool only reads but you know otherwise, mark it as changing data. You can never mark it the other way.
Turn on the tools you trust, one by one. Try it runs a tool once, as you, so you can see what it returns before any agent uses it. For a tool that changes data, that run is real.
What does "Changed by the server" mean?
When you turn a tool on, you approve that exact version: its description, its inputs and what it says about itself. Appricity checks every server's tools each day (and whenever you choose Refresh tools). If a server changes a tool you've approved, the tool turns itself off and shows Changed by the server, review it. Agents, admin chat and workflows stop using it until you read it again and choose Approve and turn on. This stops a server quietly changing what a tool does after you've trusted it.
Which agents can use which tools?
Turning a tool on makes it available; it doesn't give it to anyone yet. In Surfaces → Agents, open an agent and use its Tools tab to pick the tools that agent may use. Each tool is described to the agent on every turn, so there's a limit on how many one agent can have.
Where an agent may use them:
- In conversations with contacts (answering messages and playbooks), an agent only ever uses tools that read data. Whatever a contact writes could be an attempt to steer the agent, so a tool that can change data is never used there, whatever you pick.
- In internal tasks, an agent may also use a tool that changes data, but only if you've switched on Agents may use it in internal tasks for that tool.
- In admin chat, workspace admins can use any tool that's turned on. Before a tool that changes data runs, admin chat asks you to confirm.
Everything a tool returns is treated as information, never as instructions: an agent won't follow requests that appear inside another system's data. What an agent says to a buyer still goes through the same checks as always, including your pricing disclosure setting, whatever a tool returned.
Can a workflow call a tool?
Yes. In a workflow, Add a tool call to call any turned-on tool, with inputs filled from the event that started the workflow. Three events are especially useful for this:
- A meeting is booked: for example, create a task in your project tool.
- A conversation is escalated: for example, open a ticket in your help desk.
- An outcome is recorded: for example, post the result to your team's channel.
The result is available to later steps. If a call that changes data fails after it may have gone through, the workflow stops and says so instead of trying again, so nothing is done twice.
What about my CRM?
If you've connected your CRM through Appricity's built-in CRM integration, that connection is the only one that writes to it: leads, outcomes and status changes go through it, with your field mapping. If you also add an MCP server for the same CRM, its tools that change data stay off, so records are only ever written one way. Its read-only tools, such as looking up a deal's stage or owner in the middle of a conversation, can still be turned on. For a CRM Appricity has no built-in integration for, an MCP server is the way to connect it.
What if a server needs me to sign in again?
If the other system ends Appricity's access, the server shows Needs sign-in and workspace admins get an email. Choose Reconnect and sign in again; your tool choices are kept. You can also Turn off a server, which stops every agent, admin chat and workflow using its tools straight away, or Delete it altogether.
Which plans include it?
Connecting tools to your agents is part of the Organisation plan and above, with up to 10 connected servers on Organisation and no limit on Enterprise.