Dynatrace MCP Server

The Dynatrace MCP Server connects AI assistants such as Claude Code, Cursor, or GitHub Copilot in VS Code to the Dynatrace observability platform. Anyone monitoring applications, hosts, and Kubernetes clusters with Dynatrace can ask the assistant questions like "Which problems occurred in the last few hours?", "Why is the error rate of this service rising?", or "Which vulnerabilities affect our production environment?" without switching between the development environment and the web interface. The server is operated by Dynatrace itself. According to the documentation it is a hosted service reachable directly in your own Dynatrace environment, so nothing needs to be installed locally.

What is the Dynatrace MCP Server?

Dynatrace collects logs, metrics, traces, events, and security data and stores them in the Grail data store, which is searched with the query language DQL. The MCP server provides a collection of tools that make this data usable for external agents. According to the documentation, the address follows the pattern https://{environment-name}.apps.dynatrace.com/platform-reserved/mcp-gateway/v0.1/servers/dynatrace-mcp/mcp, and every request carries a bearer token in the header. Access therefore runs inside your own environment and within the permissions granted there.

Which use cases does it support?

The documentation names several scenarios. The assistant can use generative AI to turn a question in plain language into a DQL query, explain an existing query, and then run the generated query. It can investigate problems and vulnerabilities, analyze Kubernetes events, evaluate and forecast time series, find documents and troubleshooting guides, and resolve entity names and IDs. Translating questions into DQL in particular helps teams that do not use the query language every day.

What is it particularly useful for?

The value lies where developers need production context while they write code or investigate an incident. Instead of copying log excerpts by hand, the assistant fetches the relevant data itself, relates it to the code in the editor, and proposes causes or fixes. Compared with generic database or log servers, this server knows Dynatrace concepts such as problems, entities, and the Davis AI, and therefore returns answers that fit the monitoring environment.

Setup

For Claude Code, the official dynatrace-for-ai repository names the command claude plugin install dynatrace@claude-plugins-official; afterwards you set the environment variables DT_ENVIRONMENT and DT_PLATFORM_TOKEN. Other clients register the URL above as a remote server. According to the documentation, two ways of authenticating are available: a platform token that is passed as a bearer token and renewed manually, or an OAuth client with the authorization code flow and automatic renewal. Both need the permissions mcp-gateway:servers:invoke and mcp-gateway:servers:read; the OAuth client additionally needs scopes for AI functions, document search, DQL generation, and read access to logs, metrics, spans, events, and security data.

Security and limits

The assistant works with the rights of the token. Grant only the scopes your use case really needs and prefer read access. Keep in mind that retrieved monitoring data enters the AI client context and therefore reaches the model provider; logs can contain personal data or credentials. According to the documentation, the server currently supports neither CIMD nor public OAuth clients or dynamic client registration, which can make setup harder in some clients. The older community project dynatrace-oss/dynatrace-mcp is archived and points to this remote server.

Costs

The server itself is part of the Dynatrace platform and requires a Dynatrace environment. According to its pricing page, Dynatrace bills through a platform subscription with usage-based charges; queries on Grail can generate consumption. Which terms apply to you is shown on the official pricing page or in your contract, and a free trial environment is offered there.

Who is the Dynatrace MCP Server for?

It is worthwhile for teams that already use Dynatrace in production and want to connect their development tools with production data, such as platform, SRE, and DevOps teams. Anyone without a Dynatrace environment gains nothing from it, because the server is not a standalone monitoring solution but an access path to the platform.

A Dynatrace environment, a platform token or OAuth client with the required scopes, and an MCP client that supports remote servers, such as Claude Code, Cursor, or VS Code with GitHub Copilot.

With the Dynatrace MCP Server, an AI assistant queries problems, logs, and vulnerabilities of your Dynatrace environment and has DQL queries generated. This guide leads from creating the token to the first query.

Step 1: Check the prerequisites

You need a Dynatrace environment, an MCP client that supports remote servers, and the ability to create a platform token or an OAuth client. According to the documentation the server is a hosted service of your environment, so you do not have to install anything locally.

Step 2: Create access

According to the documentation there are two ways. A platform token is passed as a bearer token and must be renewed manually. An OAuth client with the authorization code flow renews itself automatically. Both need mcp-gateway:servers:invoke and mcp-gateway:servers:read. The OAuth client additionally needs scopes for AI functions, document search, DQL generation, and read access to storage. Grant only what you need.

Step 3: Set up in Claude Code

The official Dynatrace/dynatrace-for-ai repository names claude plugin install dynatrace@claude-plugins-official. Afterwards set DT_ENVIRONMENT to the URL of your environment, for example https://abc12345.apps.dynatrace.com, and DT_PLATFORM_TOKEN to your token.

claude plugin install dynatrace@claude-plugins-official

Other clients

Register https://{environment-name}.apps.dynatrace.com/platform-reserved/mcp-gateway/v0.1/servers/dynatrace-mcp/mcp as a remote server and pass the token in the Authorization header. Cursor and VS Code with GitHub Copilot are explicitly named in the documentation.

Step 4: First queries

Start read-only: "Which problems occurred today in the production environment?" or "Generate a DQL query for error logs of the last day and explain it." Check the generated query before having it executed.

Keep an eye on costs

Queries on Grail can generate consumption. Limit time ranges and data volumes in your questions and watch consumption in the platform.

Mind sensitive data

Retrieved logs enter the AI client context. Check whether they contain personal data or credentials before you enable the server for production environments.

Platform token as bearer header or OAuth client with the authorization code flow; both need the permissions mcp-gateway:servers:invoke and mcp-gateway:servers:read.

Read access to logs, metrics, spans, events, and security data of the environment as well as AI and document functions, within the scopes of the token used.

The server runs hosted by Dynatrace. Monitoring data the assistant retrieves enters the AI client context and therefore reaches the model provider.

Exposure of logs containing personal data or credentials to the model provider, overly broad token scopes, and unexpectedly high consumption from extensive Grail queries.