Overview
Model Context Protocol (MCP) is an open standard that enables AI agents to interact with external tools, data, and services through a single, consistent interface. Rather than building custom integrations for each tool, agents connect to MCP servers and access all the tools they expose through one standardized protocol. Learn more. Without MCP: Connecting to multiple external services requires a separate integration with custom logic for each one. With MCP: The agent communicates with all services through a single interface, dramatically reducing development complexity.Key Capabilities
- Open Standard - Universal protocol for AI and tool integration
- Simplified Development - One interface to communicate with all tools on a server
- Scalable Architecture - Add new tools without building custom integrations
- Universal Compatibility - Any MCP-compatible client works with any MCP server
When to Use
MCP tools work best when:- Tools are hosted on external servers
- You need shared toolsets across teams or multiple apps
- Tools require enterprise integrations
- You want to use third-party tool providers
- You need separation between tool logic and agent logic
| Use Case | Why MCP Works |
|---|---|
| CRM integration | Connect to a Salesforce MCP server |
| Enterprise tools | Share tools across multiple apps |
| Third-party services | Use pre-built MCP tool providers |
| Microservices | Each service exposes tools via MCP |
How MCP Works
MCP uses a client-server architecture with three components:| Component | Role |
|---|---|
| MCP Server | Hosts and exposes tools to clients |
| MCP Client | The Platform—discovers available tools and invokes them |
| MCP Protocol | Standardized communication layer (HTTP or SSE) between client and server |
Interaction Workflow
- Tool Discovery - The MCP Client connects to the server and retrieves the list of available tools.
- Intent Detection - The LLM receives the user query and the tool list, then identifies the appropriate tool.
- Tool Invocation - The client sends a structured request with the tool name and required parameters.
- Execution - The server runs the tool and returns results to the client.
- Response Generation - The agent uses the tool output to formulate a natural language response.
Example
- The Platform supports tool discovery and tool invocation from MCP servers.
- Tool definitions are fetched at configuration time and are not automatically synced. Manually reconfigure or refresh the MCP server to apply updates.
- Both HTTP and SSE transport types are supported for MCP server connections.
Configure MCP Tools
Step 1: Add an MCP Server
- Navigate to Tools → + New Tool → MCP Tool.
- Enter a Name and a Description for the MCP server.
- Under Request Definition, click Configure and provide:
- Type - Select
HTTPorSSE. - URL - The endpoint that returns tool definitions.
- Auth Profiles: - Authorization profile used to authenticate requests to the MCP server. Select None if the server requires no authentication. For OAuth, configure the OAuth settings in an Auth Profile and select it here — the platform will handle obtaining and including the access token automatically.
- Headers(Optional) - Use for add header-based authentication or custom headers required by the MCP server.
- Body - Define the request payload if the server requires additional parameters.
- Response - View the response payload sent by server.
- Type - Select
Step 2: Discover Tools
Click Test to connect to the server and retrieve its available tools. On a successful connection, the Platform displays all tools the server exposes.Step 3: Select Tools
Choose the tools to enable and click Add Selected.Step 4: Link to Agent
Add the selected MCP tools to your agent’s tool list.Manage MCP Tools
Tool Preview
The Tool Preview lets you inspect and test any MCP tool configured for your app.- Navigate to Tools within the app.
- Open the MCP tab to see all configured tools, grouped by server.
- Click any tool to view its description, input and output parameters, and sample responses.
- Tool Metadata - The name and description of the tool.
- Input Parameters - The input fields along with their data types.
-
Output Parameters - Defines how the response from the tool is processed and made available to the application. Choose one of the following modes:
- Use Complete Tool Output(Default) - Returns the entire tool response without modification. When enabled, you can include the response in the
artifactsfield of the Execute API payload using the Include Tool Response in Artifacts setting. This setting only affects the API response and does not change tool execution behavior, playground simulations, or other agent functionality. - Define Custom Parameters - Extracts specific values from the tool response and maps them as structured outputs. When selected, configure one or more output parameters by providing a name, selecting a data type, and specifying the path to the value in the response payload. Click + Add to define additional output parameters.
- Agent - Routes the value to the agent for use in subsequent processing.
- Artifacts - Stores the value in the
artifactssection of the Execute API response for programmatic access.
- Use Complete Tool Output(Default) - Returns the entire tool response without modification. When enabled, you can include the response in the
artifacts field of the Execute API response.
Tool Naming
By default, tools imported from an MCP server retain the original tool name exposed by the server when they’re added to the Agentic app. If a tool with the same name already exists in the agentic app, the system automatically prefixes the MCP server name to the tool name to avoid naming conflicts and help identify the source. Format:<MCP server name>__<tool name>.
For example, a tool named GMAIL_DELETE_DRAFT imported from a server named GoogleMCP is named as GoogleMCP__GMAIL_DELETE_DRAFT, if there is an existing tool by the name GMAIL_DELETE_DRAFT.
Testing Tools
Test tools after configuration to confirm they work correctly before using them in an agent.- Open the tool in Preview mode.
- Click Run Sample Execution.
- Enter sample values for the input parameters and click Execute.
- Review the output in the Sample Response section.
Updating Tools
When tools on the MCP server change, manually reconfigure or refresh the server to pick up the updates. The platform does not automatically sync tool changes.Refreshing the MCP Server
Click the refresh icon on the MCP server card to fetch the latest tool definitions from the server. The platform compares tools by name and updates the configuration accordingly:- If all agent-linked tools are still available on the server, the update is applied automatically without affecting the agent configuration.
- If any agent-linked tools are no longer available, the platform displays a message listing the removed/updated tools. On confirmation, the unavailable tools are unlinked from the agent and the configuration is updated with the latest available tools.
Reconfiguring the MCP Server
To edit an existing MCP server configuration:- Navigate to Tools within the app.
- Locate the MCP server card you want to update.
- Click the More options menu on the server card and select Edit Server.
- Update the required fields and click Save.
- If all app-linked tools are still available on the server, the update is applied automatically without affecting the existing tool configuration.
- If some app-linked tools are no longer available, the platform displays a message listing the affected tools. On confirmation, the affected tools are unlinked from the app.
Reference
Server Types
HTTP - Standard request/response over HTTP.Authentication
Configuration Example
Best Practices
Use Descriptive Tool Names
Write Clear Tool Descriptions
The LLM uses tool descriptions to select the right tool for a given request. Be explicit about what the tool returns and when to use it.Monitor Performance
Track these metrics to maintain tool reliability:- Response times
- Error rates
- Availability
- Token usage