The MCP Servers feature allows your organization to connect custom MCP (Model Context Protocol) servers to Cassidy, enabling your Agents to access and interact with external systems and data through standardized tool interfaces. Once an admin configures an MCP server, it can be added to any Agent, giving it the ability to perform actions defined by the server's tools directly through chat conversations.
A connector gives your Cassidy Agent the ability to take a specific action in an external system through an MCP server. Think of it like giving someone access to different functions in a custom system. Each tool exposed by the MCP server is a separate permission that you control individually.
When someone asks your Agent a question that requires data from a connected MCP server, the Agent uses these tools to query the server in real-time and get the information it needs to respond. The data is not copied or stored in Cassidy. It is only accessed when needed to answer a specific question.
For best results with complex tools, use a powerful model. Your Agent has tools that work best with these recommended models: Claude 4 Opus, Claude 4 Sonnet, GPT-5, or o3.
MCP servers can only be configured by organization admins. Once configured, they can be added to any Agent, assuming the user has permission to access that MCP server.
Setting up an MCP server (admin only)
Open Organization Settings: Click on your name in the bottom left of the sidebar, then click the settings icon for your organization.
Navigate to MCP Servers: In the Organization Settings window, click the "MCP Servers" tab on the left side of the screen.
Add a new server: Click "+ Add Server" to begin configuring a new MCP server.
Fill in server details: Complete the following fields to configure your MCP server:
"Icon" (optional): Upload an image to visually identify the server.
"Name": Enter a display name for the MCP server.
"Description": Provide a brief description of what the server does.
"Server URL": Enter the URL of your MCP server endpoint.
"Transport Type": Select "HTTP" or "SSE". We recommend HTTP for most use cases, but SSE (Server-Sent Events) is also supported.
"Authentication": Choose "OAuth 2.0" or "API Key". We recommend leaving this as OAuth 2.0 for most use cases. Some MCP servers do not use OAuth and instead require an API key, which is also supported.
"Advanced OAuth Credentials" (optional): Expand this dropdown to configure custom OAuth settings. These fields are all optional and typically only needed if you are building your own MCP servers. Most users do not need to fill these in.
"Client ID"
"Client Secret"
"Scope" (e.g., mcp.read mcp.write)
Save the server: Click "Add Server" to finish configuring the MCP server.
Share with people or groups: After adding the server, configure permissions to control who in your organization can access and use this MCP server. You can share with specific people or groups.
Your MCP server is now available to be added to new and existing Agents.
Adding an MCP server to an Agent
Navigate to Agents: Click "Agents" in the left sidebar to access your Agent dashboard.
Select your Agent: Choose the Agent you want to add the MCP server tool to by clicking on it.
Access the Connectors section: In the Agent's settings, navigate to the Connectors section of the "Setup" tab to configure and manage tools available to your Agent.
Add your MCP server: Click "+ Add Connector" and click "+ Add" next to the custom MCP server connector you want to enable for your Agent.
Choose your connection method: Select between two authentication options:
Each user connects their own account (Recommended): Each team member connects their personal account through the MCP server's authentication flow. The Agent can only access what each individual user has permission to see. This is best for teams where users have different roles and permissions.
Use a shared connection: Connect a shared service account for your entire workspace. This gives you centralized control over exactly what data the Agent can access, regardless of individual user permissions. Only workspace admins can set up shared connections. This is best for teams that want everyone accessing the same data with centralized security.
Set up shared connection (if selected): If you chose "Use a shared connection," click "Connect" to authenticate and set up the shared service account, then click "Save".
Share connection with others (optional): To share the connection with other team members, click "Change" next to the newly added connection, then click the ellipses menu (…) and select "Share".
Configure sharing permissions: Add people or groups and/or change the general access to "Team". You can set specific permissions for each person or group.
Set tool permissions: Click "Next" and then select tool permissions for each type of action. Every MCP server action is a separate tool with its own permission setting that you control individually:
Always Allow - The Agent performs this action automatically when needed
Needs Approval - The Agent asks the user for permission before taking this action
Disabled - The Agent cannot perform this action
You decide the permission level for each tool separately. When you're done configuring permissions, click "Add Tool".
Start a chat: Navigate to Chat and select the Agent with the MCP server tool enabled.
Connect individual accounts (if using "Each user" option): When first using the MCP server tool, users will be prompted to connect their account from within the chat itself.
Complete the connection: Follow the prompts to sign in with your credentials and accept the access request.
Save your settings and continue chatting: Once connected, save your settings to finalize the setup.
Your Agent can now access your MCP server's tools to help answer questions and perform actions directly through chat conversations.
Monitoring MCP Server Actions
To see what actions were taken through the MCP server, click the "Analyzed Sources" dropdown and then click on any of the MCP server steps.
If an action is called and it doesn't have the permissions, it will not allow you to perform the action.
If an action is called and needs approval, you will be able to click "Approve" or "Reject".
Data Security
Permission-based access: The Agent can only access what the connected account has permission to see.
Auditability: All queries and actions are logged in your Agent's activity history.
Service account control: Using a shared connection lets you define exactly what data and actions are accessible.