Elastic MCP server: deprecated, and what replaced it

Elastic's two MCP servers include a deprecated package with fewer tools than its documentation describes.

Last verified 1 September 2026 · from Elastic'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 Elastic MCP server?

It is Elastic's server for the Model Context Protocol. The agent gains a small set of tools for listing indices, reading mappings and running a search.

There are two surfaces and only one is developed. The standalone server, @elastic/mcp-server-elasticsearch, is the deprecated one. The Agent Builder MCP endpoint is built into Elastic 9.2.0 and later, and into Serverless projects, and is where Elastic directs users.

The standalone server is a thin wrapper over the search API. It reads; it does not write, index or administer.

Elastic's deprecated server searches Elasticsearch; its named successor is the Agent Builder endpoint in Elastic 9.2.0 and later. Both concern queried operational data, not development workflow, model traces or billing.

For an agent, Elastic answers questions you can express as a query against an index you already know exists. An agent connected here can search logs and read a mapping. It cannot list your alerts, read an APM trace or tell you what changed, so this connector is a search box rather than an observability platform.

What can an AI agent do with Elastic?

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.

This is among the cheapest connectors in the guide, and cheapness is not the point. Four tools cost almost nothing because four tools reach almost nothing. Grafana's default set costs about forty times as much and answers questions this server cannot.

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 @elastic/mcp-server-elasticsearch 0.3.1, the npm latest. It was run locally with a placeholder URL and key.

Registration is static, so no Elasticsearch instance 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 56%. 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 a small fraction.

Elastic MCP server limits: where answers come back incomplete

Two of the three limits below are about which version you are actually running, which is the recurring problem with a deprecated package.

How to configure Elastic MCP for agents

1. 1

   Check your Elasticsearch version first

   On 9.2.0 or later, the Agent Builder MCP endpoint is the supported route and this server is the bridge for older clusters.

2. 2

   Use an API key scoped to the indices you want read

   Elastic's own guidance is minimal permissions, and read-only access to specific indices where possible.

3. 3

   Configure streamable HTTP, not SSE

   The README states SSE is deprecated on this server, so a client copied from an older example may fail without a clear message.

4. 4

   Expect four tools

   The documented fifth tool, esql, runs ES|QL, Elastic's own query language. The published package does not register it, so plan around search.

Paste this into your agent’s instructions

This server returns exactly what the query asked for, so its silent failure is scope: an agent that searched one index and reports a negative answer sounds the same as one that searched everything.

When you answer using Elasticsearch MCP tools, state:
- Which indices you searched, and the time range in the query.
- The result size you requested, and whether the response hit
  it. A full page of hits is not evidence there are no more.
- Which fields you filtered on, taken from get_mappings rather
  than assumed.
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?

Often not. Elasticsearch's REST API is fully documented, and a coding agent can call it directly with the same API key. The four tools here are thin wrappers over _cat/indices, _mapping, _search and _cat/shards.

The Agent Builder endpoint is the exception and the reason to stay with MCP. It is built into the product, is where Elastic develops, and reaches further than a search wrapper.

For a cluster too old for Agent Builder, the standalone server still saves an agent from hand-rolling requests, and the cost is four tool definitions.

On Elastic 9.2.0 or later, use Agent Builder. On older clusters, the standalone server or the REST API both work, and neither is being developed further.

Elastic MCP server setup

The package runs with npx and needs ES_URL plus a credential: either ES_API_KEY, or ES_USERNAME and ES_PASSWORD.

ES_SSL_SKIP_VERIFY exists for self-signed certificates. Setting it disables certificate verification, so use it only against a cluster you control.

claude mcp add-json "elasticsearch" '{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200",
         "ES_API_KEY":"<your-api-key>"}}'

{"mcpServers":{"elasticsearch":{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200",
         "ES_API_KEY":"<your-api-key>"}}}}

[mcp_servers.elasticsearch]
command = "npx"
args = ["-y", "@elastic/mcp-server-elasticsearch"]
env = { ES_URL = "https://your-cluster:9200", ES_API_KEY = "<your-api-key>" }

Every other client

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

{"mcpServers":{"elasticsearch":{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200","ES_API_KEY":"<your-api-key>"}}}}

{"mcpServers":{"elasticsearch":{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200","ES_API_KEY":"<your-api-key>"}}}}

{"servers":{"elasticsearch":{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200","ES_API_KEY":"<your-api-key>"}}}}

{"mcpServers":{"elasticsearch":{"command":"npx",
  "args":["-y","@elastic/mcp-server-elasticsearch"],
  "env":{"ES_URL":"https://your-cluster:9200","ES_API_KEY":"<your-api-key>"}}}}

ES_URL=https://your-cluster:9200 \
ES_USERNAME=<user> ES_PASSWORD=<password> \
npx -y @elastic/mcp-server-elasticsearch

Read-only access, permissions and security

• Read-only by construction

  All four tools read. This is one of the few connectors in this guide that cannot write, without needing a flag to say so.

• Scope the API key to indices

  Elastic's guidance is minimal permissions and read-only access to specific indices where possible, which the four-tool list makes practical.

• ES_SSL_SKIP_VERIFY disables certificate verification

  It exists for self-signed certificates; using it against anything you do not control removes the transport's protection.

• Deprecated means security fixes only

  Elastic states the server will receive critical security updates and nothing else, so plan for the Agent Builder endpoint.

Troubleshooting

There is no tool for ES|QL

The published npm package registers four tools and esql is not among them. npm's latest is 0.3.1 while the repository tag is v0.4.6. Use the Agent Builder endpoint for ES|QL.

The SSE transport fails

SSE is deprecated on this server. Configure streamable HTTP, or stdio for a local process.

Certificate errors against a self-managed cluster

ES_SSL_SKIP_VERIFY bypasses verification. Prefer installing the cluster's CA where you can, because the flag removes the check entirely.

The agent cannot find a field it expects

Have it call get_mappings first. Field names come from the mapping, and a query written from memory fails silently as an empty result.

Elastic MCP server: Reference

What engineers report

Elastic's standalone MCP server has 705 stars and a clear deprecation notice, so the useful signals are versions rather than opinions.

“The get_mappings tool throws error decoding response body if a nested property is defined without explicitly specifying nested, even though the mapping is valid according to Elasticsearch specifications.”

“I deployed the MCP Server using docker and entered the container to confirm that the environment variables ES_USERNAME, ES_PASSWORD, and even ES_lOGIN configuration were all correct. However, it still said that my identity authentication failed.”

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

Should you connect an agent to Elastic?

Yes, though it is no longer actively maintained. Best for older clusters as a bridge. On 9.2.0 or later, connect Agent Builder instead.

Strongest fit

Teams on Elasticsearch clusters older than 9.2.0 who want an agent to search indices without hand-rolling REST calls.

Main advantage

It is tiny and cheap to connect, and it gives an agent the mappings it needs before writing a query.

Main weakness

Four tools reach almost nothing: no APM, no alerting, and no ES|QL on the published package.

Operational risk

Deprecated to security fixes only, and the npm package is behind the repository, so the documentation describes a server you may not be running.

Frequently asked questions