Dynatrace MCP server: tools, limits and what replaced it

Dynatrace runs three MCP servers. Four of its 18 tools act rather than read, and that is what decides how far you let an agent go.

Last verified 1 September 2026 · from Dynatrace's repository, its documentation and the server's own tools/list response

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Dynatrace MCP server?

It is Dynatrace's server for the Model Context Protocol. The agent gains tools for querying Grail, listing problems and vulnerabilities, and asking Davis AI for an analysis.

Grail is Dynatrace's data store and DQL is the query language over it. Three of the tools exist to help an agent write DQL rather than to run it: one generates a query from English, one explains a query, and one verifies a query before it runs.

There are three surfaces. This open-source server is finished at 2.1.2. `dtctl` is the successor for local development. The Remote MCP Server is hosted, generally available, and needs no local setup.

Dynatrace covers production observability broadly, reaches part of the way into release through deployment events, and reaches nothing in code review, agent observability or cost.

In practice, Dynatrace answers questions about the running system in depth, and nothing about the code that produced it or the agents you operate. An agent connected here can find a problem, query the logs around it and ask Davis AI for an analysis. It cannot see the pull request that caused it, so this connector pairs with a source-control one.

What can an AI agent do with Dynatrace?

The last row is the one to read twice. Most observability connectors only report; this one can also send. An agent that decides a problem is worth escalating has the tools to escalate it.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

Dynatrace is efficient here relative to its reach. 18 tools cost about 3,909 tokens, roughly a quarter of what Grafana's default set costs. That still covers logs, metrics, spans, problems and vulnerabilities.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted 19 August 2026 from tools/list on the published @dynatrace-oss/dynatrace-mcp-server package. It was run locally with a placeholder environment and token.

Registration is static, so no Dynatrace tenant was involved. Figures cover name, description and input schema. The two models in column one share the o200k_base encoding, so they get one count. The spread is formatting, about 31%. Reasoning tokens are output, spent per request. This count is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer. The definitions were sent to the API's count_tokens endpoint, and a no-tools baseline was subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a third.

Dynatrace MCP server limits: where answers come back incomplete

Dynatrace documents the limits below; the first is unusual enough among observability connectors to lead with.

How to configure Dynatrace MCP for agents

1. 1

   Pick the successor first

   This surface ended at 2.1.2. Dynatrace names dtctl with Dynatrace-for-AI for local development, and the hosted Remote MCP Server for remote and agent-to-agent use.

2. 2

   Decide about the four acting tools

   send_email, send_slack_message, send_event and reset_grail_budget change something outside Dynatrace or inside your quota. There is no server-side read-only flag, so the gate is your client's allowlist.

3. 3

   Set a Grail budget you can live with

   DT_GRAIL_QUERY_BUDGET_GB caps how many bytes queries may scan. Too low and exploration stops early; too high and one broad query can spend the day's budget.

4. 4

   Choose how you authenticate

   With only DT_ENVIRONMENT set, the server opens a browser for Dynatrace SSO. DT_PLATFORM_TOKEN skips that, which is what a headless setup needs.

Paste this into your agent’s instructions

Two of Dynatrace's limits produce the same empty answer: a query that found nothing, and a query that was stopped by the byte budget. The block makes the agent separate them, and makes any message it sends visible.

When you answer using Dynatrace MCP tools, state:
- The DQL you ran and the time range it covered.
- Whether the Grail query budget was reached. A stopped query
  and an empty result are not the same finding.
- Whether you sent anything. send_email, send_slack_message and
  send_event leave Dynatrace and reach people or alerting.
Never state that data does not exist. State that none was
returned under the constraints above.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Partly. The Dynatrace API covers Grail queries and problem lists, and a coding agent can call it directly with the same token.

The connector adds the DQL authoring tools, which have no API equivalent worth the name: generating a query from English, explaining one, and verifying it before it runs against a metered store.

The deprecation changes the choice rather than removing it. Between an unmaintained local package and two maintained successors, the successors win, and both keep the MCP shape.

Use dtctl locally or the Remote MCP Server for anything hosted. Call the API directly when the query is fixed and you do not need help writing DQL.

Dynatrace MCP server setup

The package runs with npx and needs DT_ENVIRONMENT, your Dynatrace environment URL. Authentication is either browser-based SSO or a platform token.

From version 1.0.0 the server no longer loads .env files automatically. Use your client's own environment configuration instead.

claude mcp add-json "dynatrace" '{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com"}}'

{"mcpServers":{"dynatrace":{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com",
         "DT_PLATFORM_TOKEN":"<your-token>"}}}}

[mcp_servers.dynatrace]
command = "npx"
args = ["-y", "@dynatrace-oss/dynatrace-mcp-server"]
env = { DT_ENVIRONMENT = "https://abc12345.apps.dynatrace.com", DT_PLATFORM_TOKEN = "<your-token>" }

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"dynatrace":{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com","DT_PLATFORM_TOKEN":"<your-token>"}}}}

{"mcpServers":{"dynatrace":{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com","DT_PLATFORM_TOKEN":"<your-token>"}}}}

{"servers":{"dynatrace":{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com","DT_PLATFORM_TOKEN":"<your-token>"}}}}

{"mcpServers":{"dynatrace":{"command":"npx",
  "args":["-y","@dynatrace-oss/dynatrace-mcp-server"],
  "env":{"DT_ENVIRONMENT":"https://abc12345.apps.dynatrace.com","DT_PLATFORM_TOKEN":"<your-token>"}}}}

DT_ENVIRONMENT=https://abc12345.apps.dynatrace.com \
DT_PLATFORM_TOKEN=<your-token> \
DT_GRAIL_QUERY_BUDGET_GB=10 \
npx -y @dynatrace-oss/dynatrace-mcp-server

Read-only access, permissions and security

• Six tools carry no read-only mark

  Four of them act on the world: two send messages to people, one ingests events that can trigger alerting, and one resets a query budget. A fifth creates a notebook. There is no server-side read-only flag.

• The token carries your platform permissions

  DT_PLATFORM_TOKEN grants whatever that token was scoped to, and the agent inherits it for every tool including the sending ones.

• Self-hosting the HTTP transport is discouraged

  Dynatrace recommends the hosted Remote MCP Server over running --http or --server yourself. Its reasons are security and no infrastructure to run.

• The server declares its own read-only tools

  12 of the 18 carry a readOnlyHint of true, so a client that respects annotations can gate on Dynatrace's own declaration.

Troubleshooting

A query returns nothing after working earlier

The Grail byte budget may be exhausted. reset_grail_budget clears the tracked count, and DT_GRAIL_QUERY_BUDGET_GB sets the ceiling.

The server opens a browser window unexpectedly

With only DT_ENVIRONMENT set, the server authenticates through Dynatrace SSO in a browser. Set DT_PLATFORM_TOKEN for a headless setup.

Environment variables from `.env` are ignored

Since version 1.0.0 the server no longer loads .env files. Use your MCP client's own environment configuration.

An unexpected email or Slack message went out

send_email and send_slack_message are in the default tool set with no read-only mode. Remove them from your client's allowlist.

Dynatrace MCP server: Reference

What engineers report

Dynatrace's connector is first-party, MIT licensed and listed on the Dynatrace Hub, with 134 stars. Its public footprint is small and its ending is clear, so the useful signals are releases rather than opinions.

“MCP Server starts, but copilot never uses the MCP server. Any prompts about the exposed tools bring nothing from the MCP server.”

Repository facts from the GitHub API, retrieved 19 August 2026. Tool counts and behavior from the server's own tools/list response on the published npm package, not from the README.

Should you connect an agent to Dynatrace?

Yes, though it is no longer actively maintained. Best for dynatrace teams connecting dtctl or the Remote MCP Server, with the four acting tools left off the allowlist.

Strongest fit

Dynatrace teams who want an agent querying Grail with help writing DQL, and Davis AI analysis in the same conversation.

Main advantage

Wide reach for a small tool list, and DQL authoring tools that make an unfamiliar query language usable by an agent.

Main weakness

This surface is finished at 2.1.2, so the choice is really between its two successors.

Operational risk

Four tools have side effects, two of which contact people. There is no server-side read-only mode.

Frequently asked questions