MCP Server
Connect external tools and services to agents using the Model Context Protocol (MCP).
What is an MCP Server?
MCP servers let agents call external tools and services, such as messaging platforms, ticketing systems, cloud APIs, and enterprise search, through the Model Context Protocol (MCP) .
The Model Context Protocol is an open standard for connecting AI applications to external data sources and tools. An MCP server exposes one or more tools - discrete operations with defined inputs and outputs that an agent can invoke during a conversation.
When a user chats with an agent, the agent can select and run MCP tools automatically based on the prompt and its configuration.
Each MCP server registers its available tools when created or synced. Chocolate Factory suffixes tool names with a server identifier to prevent collisions when multiple servers expose similarly named tools.
When to Use MCP Servers
Use MCP servers when you want agents to interact with systems outside Chocolate Factory:
- Productivity apps → send messages, create tickets, or update records in tools like Slack, Notion, or Microsoft 365.
- Enterprise search → query internal knowledge sources through connectors such as AISay or Data Xtremax.
- Custom integrations → connect proprietary APIs or internal services by registering your own MCP server URL.
- Specialized capabilities → add domain-specific tools (for example, web research, browser automation, or data warehouse queries) that are not available as built-in tools.
For native platform capabilities like code execution, chart rendering, or document generation, use built-in tools instead.
Authentication
MCP servers support several authentication methods:
| Auth type | Description |
|---|---|
| None | No credentials required. Suitable for public or locally hosted servers. |
| API Key | A shared key stored in the server configuration. |
| OAuth | Redirects to the provider for authorization using OAuth protocol. |
| Custom | A JSON object of headers or credentials forwarded to the MCP server. |
Account-Level Authentication
Some MCP servers require per-user credentials rather than shared organization credentials. Enable Use account-level authentication on the server detail page to load auth from each user’s profile instead of the server-level configuration.
When an agent uses a server with account-level auth, users are prompted to configure their credentials from the Dashboard before chatting. Click Configure Credentials in the alert banner to complete setup.
OAuth and API key credentials are sensitive. Only users with appropriate permissions can create or update MCP server configurations. Never share credentials in chat messages or public repositories.
Tips for Effective Prompts
To help your agent choose the right MCP tools, be explicit about the tool and the action you want:
- Name the service → “Search AISay for the latest policy on remote work” instead of “find the remote work policy.”
- Specify the action → “Create a Jira ticket in the ENG project” instead of “log this bug.”
- Provide required inputs → include URLs, IDs, channel names, or other parameters the tool needs.
- State the expected output → “Summarize the top three results” or “Post the summary to #team-updates.”
When an agent uses an MCP tool, you will see an inline indicator in the chat showing the tool name and status. Expand the reasoning trail to review each tool call, its inputs, and the results.
See Chat with Agents for more guidance on writing effective prompts and reviewing tool activity.
Ready to connect your first MCP Server? See Setup & Configuration.