MCP and connecting DockRay to AI assistants

DockRay provides an MCP server - a standard way to connect tools to AI assistants. This lets an assistant use monitoring data in a controlled way, without copying alerts and statistics between applications.

What does an MCP connection give you?

An AI assistant can read the list of projects, their current state, recent alerts, aggregated statistics, and basic information about error groups. This makes it faster to gather the context needed for diagnosis and lets the team focus on solving the problem.

Once additional permissions are granted, the assistant can also create and configure projects, change the status of errors, or delete selected resources. Each key has its own access scope, so you don't have to expose every operation.

How can AI help in practice?

• compare a rise in error counts with recent alerts and project statistics;
• prepare an incident summary for a developer or the person responsible for the project;
• highlight projects that need attention and prioritise the order of diagnosis;
• change the status of an analysed error group, if the key has that permission;
• handle a repeatable process, for example creating a project and setting up monitoring according to an agreed standard.

In a more advanced process, an assistant can combine information from DockRay with a code repository, an issue tracker, and documentation. It can then analyse the problem, point to a likely cause, propose a fix, and prepare it for review. MCP gives it the current operational context, which the AI model doesn't have on its own.

Setting up the connection

1. In the panel, go to the MCP integration settings.
2. Create a separate key for a specific assistant or process.
3. Choose only the permissions needed for that use case.
4. Copy the token and store it in an environment variable of your MCP client. The full value is shown only once.
5. Add the endpoint address and token to the configuration of the MCP client you're using.

Don't store the token in a repository, in a file shared with the team, or in the content of a prompt sent to the model. If a key is no longer needed or may have been exposed, revoke it and create a new one.

Automation needs control

AI can speed up analysis and handle part of the repetitive work, but it shouldn't make irreversible decisions on its own. Start with read-only permissions. Grant write or delete access only when the process has clearly defined rules, a limited scope, and a person responsible for oversight.

Proposed changes to the application should go through tests and standard code review. Before deployment, a human should confirm the cause of the problem, assess the impact of the fix, and check that it doesn't compromise security, customer data, or other parts of the system.

The safest way of working is automatic analysis and preparation of an action, combined with human approval before any change with significant consequences.