New Relic MCP server (Ground Truth): 38 tools and setup

New Relic's hosted observability connector, now part of Ground Truth: its three regional endpoints, its read tools, and the ones that need OAuth.

Checked 1 September 2026 · from New Relic's own MCP documentation and tool reference

One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.

What is the New Relic MCP server?

It is New Relic's server for the Model Context Protocol. New Relic is an observability platform, so a connected agent reaches telemetry, alerts and the entities that produce them.

Two tools carry the querying. execute_nrql_query runs NRQL directly, and natural_language_to_nrql_query turns a question into NRQL. The pairing is rare. The agent can write the query itself, or it can ask New Relic to write it.

The rest is analysis rather than retrieval. analyze_deployment_impact, analyze_golden_metrics, analyze_threads and generate_user_impact_report return New Relic's own assessment rather than raw rows.

It is strongest in production observability, touches release through change events, and reaches nothing in code review or cost.

For an agent, New Relic answers what production is doing and what changed just before it started. Deployment impact analysis is the tool that links the two. It has no view of the code, so a fix still needs a repository connector.

What can an AI agent do with New Relic?

list_available_new_relic_accounts belongs first. Most tools take an account, and an answer from the wrong one looks exactly like an answer from the right one.

What the tool definitions cost in context

A client that loads every tool definition at the start of a session spends this much context before the first question. Claude Code defers definitions by default through tool search, and loads a tool's definition when the agent needs it. The method below gives the version and date measured.

A context window is the amount of text a model can hold at once. The 27 tools documented on 24 August 2026 cost 1,897 to 2,193 tokens on Claude. New Relic documents 38 as of 6 October 2026, so the current list costs more. The include-tags header filters tools by tag and trims that cost.

New Relic's list grew from 27 to 38 tools between August and October 2026, so the surface that registers changes and is not one to budget against.

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 from the 27 tool entries in the vendor's own documentation on 24 August 2026. That is the only public inventory.

The server could not be run for a live tools/list. So no input schemas are available. These figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint, with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. 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 fifth.

New Relic MCP server limits: where answers come back incomplete

New Relic's authentication split, regional deployments and generated queries set the constraints below.

How to configure New Relic MCP for agents

1. 1

   Check access first

   An administrator turns MCP on in feature control manager. The server and local-client toggles are on by default. API key access and preview tools stay off until an administrator enables them.

2. 2

   Choose OAuth if you need the AI tools

   natural_language_to_nrql_query, analyze_deployment_impact, the two report generators and the three preview tools work only over OAuth.

3. 3

   Pick the region

   Endpoints differ: mcp.newrelic.com for the US, mcp.eu.newrelic.com for the EU and mcp.jp.newrelic.com for Japan.

4. 4

   Decide between a user key and OAuth

   A user key is in the NRAK- format; OAuth uses New Relic's pre-configured client and endpoints.

5. 5

   Trim the tool list with tags

   The include-tags header limits tools to New Relic's tags, for example discovery,alerting.

Paste this into your agent’s instructions

Several tools generate an assessment rather than returning data, and one writes the query the answer rests on. Both need naming.

When you answer using New Relic MCP tools, state:
- The account and entity the answer covers.
- The NRQL that ran. If natural_language_to_nrql_query wrote
  it, show the query it produced.
- Whether a claim came from an analyze_ or generate_ tool,
  which return New Relic's assessment rather than raw telemetry.
Never state that data does not exist. State that the query
above returned no results.
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?

Often not. NerdGraph and the NRQL API answer the same questions, and a coding agent can call them directly.

The connector earns its place on the analysis tools. Deployment impact and golden-metric analysis are New Relic's own assessments, which a raw query does not reproduce.

Grafana and Datadog both publish generally available servers over their own telemetry.

Connect it for investigation. Use NerdGraph for scripted or bulk pulls.

New Relic MCP server setup

The server is hosted, so nothing is installed. An administrator enables it in feature control manager, and your group needs an organization role that can read the MCP server.

Authentication is a user key or OAuth. Keys are in the NRAK- format and are created under API keys in the account menu. With a key, the four AI analysis tools and the three preview tools are unavailable; they need OAuth.

claude mcp add --transport http newrelic \
  https://mcp.newrelic.com/mcp \
  --header "Api-Key: NRAK-XXXXXXXXXXXXXXXXXXXXX"

{"mcpServers":{"newrelic":{
  "url":"https://mcp.newrelic.com/mcp",
  "headers":{"Api-Key":"NRAK-XXXXXXXXXXXXXXXXXXXXX"}}}}

[mcp_servers.newrelic]
url = "https://mcp.newrelic.com/mcp"
headers = { "Api-Key" = "NRAK-XXXXXXXXXXXXXXXXXXXXX" }

Every other client

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

{"mcpServers":{"newrelic":{"url":"https://mcp.newrelic.com/mcp",
  "headers":{"Api-Key":"NRAK-XXXXXXXXXXXXXXXXXXXXX"}}}}

{"mcpServers":{"newrelic":{"url":"https://mcp.newrelic.com/mcp",
  "headers":{"Api-Key":"NRAK-XXXXXXXXXXXXXXXXXXXXX"}}}}

{"servers":{"newrelic":{"url":"https://mcp.newrelic.com/mcp",
  "headers":{"Api-Key":"NRAK-XXXXXXXXXXXXXXXXXXXXX"}}}}

{"mcpServers":{"newrelic":{"url":"https://mcp.newrelic.com/mcp",
  "headers":{"Api-Key":"NRAK-XXXXXXXXXXXXXXXXXXXXX"}}}}

# EU
https://mcp.eu.newrelic.com/mcp
# Japan
https://mcp.jp.newrelic.com/mcp

# Authorization URL: https://oauth2.service.newrelic.com/oauth2/auth
# Token URL:         https://oauth2.service.newrelic.com/oauth2/token
# Scopes:            ["observability:read", "offline", "offline_access"]

Read-only access, permissions and security

• Check compliance with New Relic directly

  The preview documentation prohibited FedRAMP-regulated accounts. The current MCP documentation, read 6 October 2026, makes no FedRAMP statement.

• Three tools are still preview

  search_traces, analyze_trace_anomaly and analyze_service_dependencies are read-only and provided for evaluation.

• Telemetry can carry sensitive content

  Log and transaction tools return production data, which reaches the model like any other tool result.

• Every documented tool reads

  All 38 query, list, analyze or report, and New Relic labels the preview tools read-only.

Troubleshooting

The endpoint refuses the connection

An administrator has to enable MCP in feature control manager, and your group needs an organization-scoped role that can read the MCP server. API key connections also need the API Keys toggle.

Data appears to be missing

Check the region and the account. There are separate US, EU and Japan deployments, and list_available_new_relic_accounts shows what the key can reach.

A natural-language answer looks wrong

natural_language_to_nrql_query writes the query, and execute_nrql_query runs it. Have the agent print the generated NRQL and read it.

Some documented tools are missing

The four AI analysis tools and the three preview tools are unavailable with a user key; connect over OAuth. Preview tools also need the Preview Tools toggle.

New Relic MCP server: Reference

What engineers report

New Relic documents 38 tools as of 6 October 2026 and runs three regional endpoints, and the figures below come from those docs. Public discussion is small: newrelic/mcp-server has 7 stars, and the two issues below are what engineers have filed against it.

“The MCP tool is great but Agents will run the same query across all accounts for every investigation unless specifically prompted not to do so.”

“Even when I am already logged in, when clicking on the MCP oauth link, I am redirected out to my corporate SSO. After login, the mcp oath flow was lost, and nothing ever goes back to the MCP.”

Tool inventory, regions, authentication and status read from New Relic's own MCP documentation and tool reference on 6 October 2026. Repository facts from the GitHub API on 24 August 2026.

Should you connect an agent to New Relic?

Yes. Best for OAuth in the right region, with generated NRQL printed in every answer.

Strongest fit

New Relic customers who want an agent to investigate production from the editor.

Main advantage

Both routes to a query: the agent can write NRQL, or ask the platform to write it.

Main weakness

The AI analysis tools need OAuth and the Advanced Compute add-on, so a key-based setup loses New Relic's own assessments.

Operational risk

Low from the tools, which all read. The main risk is a generated NRQL query that measures the wrong thing.

Frequently asked questions