Elastic Agent Builder MCP Server for Claude

Content

Elastic Agent Builder MCP Server is Elastic's official MCP endpoint for connecting external AI clients to Elastic Agent Builder tools. Instead of running a standalone Elasticsearch MCP package, you connect the client to the Kibana endpoint for the deployment and authenticate with an Elastic API key scoped to the exact indices, Kibana space, and Agent Builder privileges the assistant should use.

Use it when Claude needs a governed bridge into Elastic Agent Builder workflows for logs, metrics, search, observability, security, or document exploration. The useful part is also the risky part: the assistant sees whatever the API key and configured Agent Builder tools can access, so the right default is a short-lived, least-privilege API key tied to a narrow space and index pattern.

Features

• Official Elastic Agent Builder MCP endpoint documented by Elastic.
• Standard endpoint at {KIBANA_URL}/api/agent_builder/mcp.
• Custom Kibana space endpoint at {KIBANA_URL}/s/{SPACE_NAME}/api/agent_builder/mcp.
• Setup pattern for MCP clients such as Claude Desktop, Cursor, VS Code, and similar clients.
• Uses mcp-remote to bridge local MCP clients to the Kibana HTTP endpoint.
• API key authentication through an Authorization: ApiKey ... header.
• Kibana application privilege requirements for Elastic Agent Builder.
• Elasticsearch index privilege scoping through role descriptors.
• Support for narrow production API keys that expose only selected index patterns such as logs-* or metrics-*.
• Explicit docs guidance to set API key expiration dates and rotate keys.
• Works with the Agent Builder tools configured in the target Elastic deployment and Kibana space.

Use Cases

• Let Claude inspect approved Elastic logs or metrics indexes during incident triage without giving it broad cluster access.
• Expose a narrow Agent Builder toolset for support investigations in one Kibana space.
• Query selected observability indices and summarize recurring errors, affected services, or recent patterns.
• Use an API key scoped to logs-* and metrics-* while keeping sensitive security or customer-data indices out of the assistant's reach.
• Test Agent Builder tools from Claude Desktop, Cursor, or VS Code before publishing them to a wider team.
• Reproduce a 403 by checking whether the API key includes the required Kibana Agent Builder application privileges.

Installation

Default Kibana space

Set the Kibana URL and API key:

export KIBANA_URL="https://your-deployment.kb.us-central1.gcp.cloud.es.io"
export API_KEY="your-api-key"

Then configure the MCP client with mcp-remote:

{
  "mcpServers": {
    "elastic-agent-builder": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "${KIBANA_URL}/api/agent_builder/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "KIBANA_URL": "${KIBANA_URL}",
        "AUTH_HEADER": "ApiKey ${API_KEY}"
      }
    }
  }
}

Custom Kibana space

For a non-default Kibana space, include the space segment:

{
  "mcpServers": {
    "elastic-agent-builder": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "${KIBANA_URL}/s/${SPACE_NAME}/api/agent_builder/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "KIBANA_URL": "${KIBANA_URL}",
        "SPACE_NAME": "default",
        "AUTH_HEADER": "ApiKey ${API_KEY}"
      }
    }
  }
}

API key shape

Elastic's docs show an API key role descriptor with cluster access for inference monitoring, index-level read and view_index_metadata, and Kibana application privileges for Agent Builder. For production, restrict index names to the smallest useful patterns:

{
  "indices": [
    {
      "names": ["logs-*", "metrics-*"],
      "privileges": ["read", "view_index_metadata"]
    }
  ],
  "applications": [
    {
      "application": "kibana-.kibana",
      "privileges": ["feature_agentBuilder.read", "feature_actions.read"],
      "resources": ["space:default"]
    }
  ]
}

Examples

Inspect approved logs

Use Elastic Agent Builder MCP to summarize recurring errors in the approved logs-* indices from the last 30 minutes. Do not change detections, dashboards, or workflows.

Verify MCP permissions

Before querying data, confirm which Kibana space and index patterns this Elastic MCP connection can access.

Debug 403 access

Use the Elastic MCP setup context to explain why the current API key might receive 403 Forbidden when connecting to the Agent Builder endpoint.

Keep scope narrow

Only use Elastic Agent Builder MCP tools that read logs and metrics. Avoid security, customer-data, or broad wildcard index queries.

Source Notes

• Elastic's official docs describe the MCP server as a standardized interface for external clients to access Elastic Agent Builder tools.
• Elastic documents the endpoint as {KIBANA_URL}/api/agent_builder/mcp and the custom-space variant as {KIBANA_URL}/s/{SPACE_NAME}/api/agent_builder/mcp.
• Elastic's client example uses npx mcp-remote with an API key authorization header.
• Elastic documents required Kibana application privileges for Agent Builder and notes that missing feature_agentBuilder.read causes a 403 Forbidden connection failure.
• Elastic's best practices recommend API key expiration dates and limiting API keys to only the indices the tools need to access.

Duplicate Check

Checked current upstream/main, open PR titles, open PR changed files, source URLs, and content files for Elastic Agent Builder MCP, Elastic MCP, Elasticsearch MCP, elastic-agent-builder-mcp-server, /api/agent_builder/mcp, and elastic.co/docs/explore-analyze/ai-features/agent-builder/mcp-server. Existing content mentions Elastic only in broad infrastructure, search, or observability examples; there is no dedicated Elastic Agent Builder MCP Server entry, source URL duplicate, or open content PR for this server.

Editorial Disclosure

Elastic is a commercial search, observability, and security platform with open-source components, but this listing is not sponsored, paid, affiliate-backed, or submitted by Elastic. Use Elastic's current Agent Builder docs, Kibana space model, API key privileges, and deployment policies as the source of truth before connecting sensitive indices to any AI client.