PlugWiki

Elastic Agent Builder MCP Server (Kibana)

OfficialRemote

Lets Claude, Cursor or ChatGPT search and query your Elasticsearch data through Kibana, with no extra server to run. Elastic 9.3+ or Serverless.

This is Elastic's official way to connect AI assistants to your Elasticsearch data, and it is built into Kibana. Your assistant can list indices, look at mappings, search in plain language and write and run ES|QL queries, plus any custom tools your team adds in Agent Builder. It sees whatever the API key or signed-in user can see, so give it a key limited to the indices you want to share. It needs Elastic 9.3 or later (preview in 9.2) or a Serverless project, and replaces Elastic's older standalone server, which now gets only security fixes.

Good for

  • Elastic users who want to ask Claude or Cursor about their logs, metrics or search indices
  • SREs and analysts who want ES|QL queries written and run from chat
  • Teams on Elastic Cloud Serverless who want people to sign in with their own Elastic account
  • Anyone still on Elastic's old standalone Elasticsearch MCP server

Works in

5 supported · 8 should work

Tap an app to see how to set it up and any known problems.

Install Elastic Agent Builder MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Elastic Agent Builder MCP server to Claude Code. Run this command:
    
    claude mcp add --transport http --client-id {CLIENT_ID} kibana-mcp {MCP_SERVER_URL}
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Elastic Cloud Serverless project, OAuth client registered in the project.

Read-only mode: Built-in tools only read. To keep it that way, sign in as an Elastic user whose role reads only the indices you need, and avoid adding custom tools that change data.

Run /mcp or claude mcp login kibana-mcp to open the sign-in page. For a confidential client add --client-secret (Claude Code prompts for it). Elastic documents Claude Code only for Serverless sign-in.

From the vendor's docs. Source.

Tools

  • Searches your Elasticsearch data from a plain-language question

    Searches your indices from a question in plain words, choosing the right kind of query for you.

    Try: “Find error logs from the checkout service in the last hour”

    Read · platform_core_search

  • Runs an ES|QL query and returns a table

    Runs an ES|QL query, one you wrote or one the assistant generated, and returns the results as a table.

    Try: “Run FROM logs-* | STATS count() BY host.name and show the top 10”

    Read · platform_core_execute_esql

  • Writes an ES|QL query from a plain-language request

    Turns a request in plain words into an ES|QL query you can run or reuse.

    Try: “Write an ES|QL query for average response time per service today”

    Read · platform_core_generate_esql

  • Lists the indices, aliases and data streams you can access

    Lists the indices, aliases and data streams in your cluster that your key or account can see.

    Try: “Which indices do I have?”

    Read · platform_core_list_indices

  • Finds which indices hold the data you are asking about

    Suggests the indices most likely to hold the data you describe, with their field layouts.

    Try: “Where are our payment events stored?”

    Read · platform_core_index_explorer

  • Opens one document by its ID

    Gets the full content of one document, given its ID and index.

    Try: “Show me document abc123 from the orders index”

    Read · platform_core_get_document_by_id

  • Looks up Elastic's product documentation

    Searches Elastic's documentation, if it is installed in your deployment, to answer how-to questions.

    Try: “How do I set up index lifecycle policies?”

    Read · platform_core_product_documentation

  • Answers questions about your installed data integrations

    Searches what your installed Fleet integrations document, such as how to configure them to bring data in.

    Try: “How is our AWS CloudTrail integration configured to collect logs?”

    Read · platform_core_integration_knowledge

  • Used by the assistant behind the scenes

    Shows the fields and types in an index

    Gets the mapping of one or more indices, so the assistant knows which fields exist and their types before it queries.

    Try: “What fields does the logs-nginx index have?”

    Read · platform_core_get_index_mapping

From the official docs

Pros and cons

Good

  • Official and built into Kibana, with nothing extra to host.
  • Search, index lookup and ES|QL tools work on your live data.
  • Custom tools and workflows you build in Agent Builder show up too.
  • Serverless users can sign in with their own Elastic account and permissions.
  • Listed in Claude's connectors directory for Serverless projects.
  • Access follows the API key or user, so you can limit it to certain indices.

Watch out

  • Needs Elastic 9.3 or later, or Serverless; 9.2 is preview only.
  • Requires a paid Elastic subscription or Serverless feature tier.
  • Sign-in is Serverless only; Cloud Hosted and self-managed need an API key.
  • Desktop apps need the mcp-remote helper when you use an API key.
  • You cannot limit which tools a key can call.
  • Large results can break some MCP clients.

Permissions

low risk · 3 things to watch

Low risk

The built-in tools only read, so the main risk is your assistant seeing more data than you meant to share. Use an API key limited to the indices you need, with an expiry date.

What it can do

Watch out for

  • Any key with access can use every built-in tool

    You cannot yet limit which tools a key or role may call. Limit the key to specific indices instead. github.com

  • An API key is one shared identity

    Everyone using the same key acts with the same permissions. Set an expiry, 1 to 7 days for testing and 30 to 90 for production, and rotate it. elastic.co

  • Custom tools can do more than read

    Custom tools and Elastic Workflows you add in Agent Builder are exposed too. Check what each one does before connecting. elastic.co

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyBuilt-in core tools only read. Create the API key with read and view_index_metadata index privileges on the indices you want exposed, or authorize OAuth with a limited user.
Sign-in and keysAPI key in your MCP client config (as an Authorization header), or short-lived OAuth tokens on Serverless.
Scopes
  • Indices and data the API key or signed-in user can read

  • Kibana feature privilege feature_agentBuilder.read (feature_actions.read in Elastic's example key)

  • Custom Agent Builder tools and Elastic Workflows in the space

Alternatives

Use the Agent Builder endpoint if you run Elastic 9.3 or later or Serverless. On older clusters, a community server that talks to Elasticsearch directly is the fallback, since Elastic's own standalone server is deprecated.

ServerOfficialRunsSign-inStatusGitHub stars
For Kibana
Elastic Agent Builder MCPThis pageOfficialIn KibanaAPI key or accountActive–
mcp-server-kibanaCommunityLocal––85
Other tools for the same job
Elasticsearch MCP ServerOfficialDockerAPI keyDeprecated728
elasticsearch-mcp-serverCommunityLocal––314

What people say

People like that it is built into Kibana, and teams ask for finer control over which tools a key can use.

What people like

  • Built into Kibana, replacing the separate server Elastic used to ship

Common complaints

  • No way to limit which tools an API key or role can use
  • Built-in tools cannot be hidden from the MCP server
  • Large results arrive as one long line and fail in some clients

In their words

Questions

What is the Kibana MCP server URL?

{KIBANA_URL}/api/agent_builder/mcp. For a custom space, use {KIBANA_URL}/s/{SPACE_NAME}/api/agent_builder/mcp. You can also copy it from the Tools page in Agent Builder.

Which Elastic versions support it?

Elastic Cloud Serverless, and Elastic Stack 9.3 or later. It was a preview in 9.2.

Why do I get 403 Forbidden?

Your API key is missing the Agent Builder read privilege in Kibana (feature_agentBuilder.read). Create a new key that includes it.

Can I sign in instead of using an API key?

Only on Serverless, through an OAuth client you register in the project. Elastic Cloud Hosted and self-managed clusters use API keys.

Is it free?

No. Agent Builder needs the right Elastic Stack subscription or Serverless project feature tier.

Can it change or delete my data?

The built-in tools only read. Custom tools or workflows your team adds in Agent Builder can do more, so check them before connecting.

Why does it ask me to sign in again?

On Serverless, a sign-in expires after 30 days without use. Sign in again to reconnect.

Is Elastic's old Elasticsearch MCP server still supported?

It is deprecated and gets only critical security updates. Elastic points users to this endpoint instead.

Elastic Agent Builder MCP vs the Elasticsearch MCP server?

Agent Builder runs inside Kibana 9.3+ with no extra server and exposes your custom tools too. The old server runs as a separate container and works with 8.x clusters, but is no longer developed.

Examples

  • Run an ES|QL query from Claude Desktop without opening Kibana

    Source
  • Explore your indices and their mappings

    Source
  • Investigate logs and other Observability data

    Source
  • Run and monitor Elastic Workflows

    Source

Information

updated Nov 1, 2026

Publisher
Official
Transport
Streamable HTTP
Auth
API key

DocsClaude connectors directory (Serverless)Connect with an API key

Other Kibana MCP servers

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 10
  1. Connect an MCP host to Elastic Agent Builder · elastic.co
  2. Authenticate MCP clients with API keys · elastic.co
  3. Elastic Cloud Serverless connector for Claude · elastic.co
  4. Elastic Agent Builder MCP server · elastic.co
  5. Elastic Agent Builder built-in tools reference · elastic.co
  6. elastic/mcp-server-elasticsearch · github.com
  7. MCP tool responses are serialized as a single very long JSON line · github.com
  8. Restrict MCP server tools per role / API key · github.com
  9. Elastic Agent Builder · elastic.co
  10. Allow excluding specific built-in tools from Agent Builder MCP exposure · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 1, 2026.