generated: '2026-08-29'
method: probed
source: https://docs.streamsec.io/docs/stream-security-mcp-server-preview
status: published
name: Stream Security MCP Server
description: >-
  Lets an agent interact with Stream Security data in natural language - retrieving resource
  metadata, configuration changes, misconfigurations, external exposures, excessive privileges and
  access to critical data, and triaging or investigating threat detections against identity,
  network and Kubernetes audit logs. Documented as a preview.
deployment:
  mode: remote
  endpoint: https://app.streamsec.io/mcp
  auth: api-key
  verified: probed
  note: >-
    mode is `remote`, not local-stdio. The provider's own configuration block runs
    `npx -y mcp-remote https://app.streamsec.io/mcp` - mcp-remote is a generic third-party stdio
    BRIDGE, not a Stream Security package, and the thing an MCP client actually talks to is the
    https endpoint. Stream.Security publishes no MCP server package of its own to npm or PyPI
    (both probed, 0 results), so there is no local-stdio product here.
x-evidence:
  fetched: '2026-08-29'
  url: https://app.streamsec.io/mcp
  method: POST
  request: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
  http_status: 200
  content_type: application/json+stream; charset=utf-8
  response: '{"jsonrpc":"2.0","id":1,"error":{"code":-32603,"message":"Valid Bearer token required for MCP access"}}'
  interpretation: >-
    The endpoint is LIVE and is a real MCP JSON-RPC server - it parsed the tools/list envelope and
    answered with a well-formed JSON-RPC error rather than an HTML 404. The tool schemas are
    auth-gated behind a Stream Security API token.
authentication:
  type: api-key
  transport: Authorization header, carried through by mcp-remote --header "Authorization:$STREAMSEC_AUTH"
  workspace_header: 'customer: <WORKSPACE_ID>'
  token_source: Organization/Workspace Settings -> API Token Management -> Create API Token
  docs: https://docs.streamsec.io/docs/token-management
  note: >-
    One workspace per API token. Multiple workspaces require a separately configured MCP server
    entry per workspace, each with its own token and `customer` header.
tools:
  status: gated
  count: null
  note: >-
    NOT ENUMERATED. tools/list requires a Bearer token, and the provider does not publish a tool
    list in its documentation or in llms.txt - the MCP page describes capabilities in prose only.
    No tool names are recorded here because none are published; inventing them would fabricate an
    agent surface. Authenticated introspection is required to obtain the real names and
    inputSchemas.
documented_capabilities:
- Retrieve environment insights - resource metadata, configuration changes, misconfigurations,
  external exposures, excessive privileges, access to critical data
- Threat investigation - triage and investigate detections using investigation data combined with
  identity, network and Kubernetes audit logs
clients_documented:
- Claude Desktop
- Cursor IDE
requirements:
- Node.js installed (for the mcp-remote bridge)
- A Stream Security API token