Model Context Protocol (MCP) Integration
Model Context Protocol (MCP) is an open-source standard for connecting AI applications to external systems. MCP integrations are used for custom-built tools that do not have a standard integration with Biggy.
MCP integrations can be used with the Observability Agent action plan, or as an ITSM tool.
To create an MCP integration:
On the web app integrations page, click + Add MCP.
Select the Service Type. You can choose from:
Observability: Monitoring, metrics, logs, and traces from observability platforms.
ITSM: IT service management for incident tracking and change workflows.
Context Retrieval: Retrieval-only sources (runbooks, post-mortems, knowledge base content) that Biggy can query for troubleshooting context.
Other: Any other MCP server. Describe what it does, and Biggy will route requests to it based on your description.
Enter a descriptive Name for the MCP.
In the configuration page, populate the following information:
Field
Description
Authentication Method
Select the authentication method that should be used for this MCP server. The following options are available:
Bearer Token: Sent as "Authorization: Bearer <token>". The standard for authenticated MCP servers.
API Key Header: Sent as a request header you name (For example,
X-API-Key: <key>.). For MCPs that expect an API key instead of a bearer token.Basic Auth: Sent as
Authorization: Basicwith a username and password. Rare for MCP servers.OAuth 2.0 (Client Credentials): A short-lived bearer token is obtained from your Token URL using the client credentials grant, then sent in the Authorization header as
Authorization: Bearer <token>.Tokens are refreshed automatically before they expire. With a relay client assigned, the Token URL is reached from inside your network; otherwise, it must be publicly resolvable.
OAuth 2.0 (Authorization Code): An admin authorizes Biggy once at your identity provider. Biggy stores the refresh token and uses it to obtain short-lived bearer tokens, sent as
Authorization: Bearer <token>. Leave the endpoints blank to discover them from the MCP server, and the Client ID blank to register Biggy automatically when the provider supports it.No Authentication: No credentials are sent. This method is typically used only for MCP servers running on your network and accessed via a relay client.
MCP Server URL
Base URL of your MCP server. Include the port if needed.
Authentication
Depending on the selected authentication method, enter your MCP server credentials.
Custom Headers (Optional)
Add custom HTTP headers to include with all API requests. For each header, include the Header Name and Header Value. To add additional headers, click the + sign.
On-Premises Instance
Enable when this MCP instance is only reachable from inside your network. Requests route through a relay client, assigned on the Relay Clients page.
Response Time
Define how long Biggy waits for the server to answer a tool time.
In the Tool call timeout field, select a time in seconds.
Leave blank to use the 60-second default. Raise it only for servers whose queries take longer. The maximum is 5 minutes. Other MCP servers are not affected.
Requester Identity
Enable Share the requesting user's email to allow the server to see which Biggy user is behind the chat request.
How Biggy understands this server
Enter Routing guidance to help Biggy understand when to use this server. Each part pairs the server's self-description with what you write for your environment. When they disagree, your description takes precedence.
Agent Notes
List any notes or guidelines to help Biggy understand your MCP instance. For example, if your logs or alerts contain a specific field used to identify the type of application or service, provide that information here.
We highly recommend configuring this field, as it lets you enter organization-specific information that helps Biggy provide more accurate and consistent results.
Allowed MCP Tools
Choose which tools this MCP server can run on your behalf.
In the Trust Policy section, choose how Biggy decides which tools are callable. Select from the following options:
Allow only selected tools (Recommended): Pick exactly which tools Biggy can call. New tools require explicit approval.
When this option is selected, use the search bar to select one or more tools.
Auto-trust all tools: Allow every tool this server exposes, including tools added in the future.
Human Approval
Decide which tools require explicit human approval before Biggy runs them. The following policies are available:
Destructive only: Default. Only tools that delete or destroy data ask for approval.
All writes: Ask before any tool that writes, modifies, or deletes. Read-only tools still run automatically.
Strict: Gate writes and deletes, plus unclassified tools whose effect can't be determined from their name.
Off: Never ask. Only for fully-trusted servers. Every tool allowed runs without approval.
Click Save Instance.