generated: '2026-09-01'
method: searched
status: published
source: >-
  https://awslabs.github.io/mcp/ and https://github.com/awslabs/mcp — plus a live anonymous
  JSON-RPC probe of https://knowledge-mcp.global.api.aws/mcp on 2026-09-01.
deployment:
  mode: both
  endpoint: https://knowledge-mcp.global.api.aws/mcp
  install: uvx awslabs.aws-api-mcp-server@latest
  package: https://pypi.org/project/awslabs.aws-api-mcp-server/
  auth: none
  verified: probed
note: >-
  AWS ships BOTH shapes of MCP server and they are not the same product. The AWS Knowledge MCP Server is
  a hosted remote endpoint an agent can POST to right now with no account, no key and no install — an
  `initialize` and a `tools/list` both returned HTTP 200 anonymously during this pass. Separately,
  github.com/awslabs/mcp publishes 62 first-party local-stdio servers (one per service area) distributed
  through PyPI under the `awslabs.*` namespace; those require a human to install and run them on a
  machine before any agent sees them. Both are recorded because both are real work by the provider, and
  only the first is reachable by an agent unattended.
servers:
- name: AWS Knowledge MCP Server
  kind: remote
  transport: streamable-http
  url: https://knowledge-mcp.global.api.aws/mcp
  auth: none
  server_info:
    name: AWSKnowledgeMCP
    version: 1.0.0
    protocolVersion: '2025-03-26'
  docs: https://github.com/awslabs/mcp/tree/main/src/aws-knowledge-mcp-server
  tools_file: amazon-web-services-aws-mcp-tools.json
  x-evidence:
    probed: '2026-09-01'
    initialize_status: 200
    tools_list_status: 200
    tool_count: 5
- name: awslabs MCP server suite
  kind: local-stdio
  transport: stdio
  install: uvx awslabs.<server-name>@latest
  registry: pypi
  namespace: awslabs.*
  repository: https://github.com/awslabs/mcp
  docs: https://awslabs.github.io/mcp/
  server_count: 62
  auth: aws-credentials
  note: >-
    Runs under the operator's own AWS credentials (SigV4 via the standard credential chain), so its
    reach is whatever the caller's IAM principal is allowed to do.
  examples:
  - aws-api-mcp-server
  - aws-documentation-mcp-server
  - aws-knowledge-mcp-server
  - aws-serverless-mcp-server
  - aws-pricing-mcp-server
  - billing-cost-management-mcp-server
  - cloudwatch-mcp-server
  - dynamodb-mcp-server
  - ecs-mcp-server
  - eks-mcp-server
  - iam-mcp-server
  - lambda-tool-mcp-server
  - postgres-mcp-server
  - s3-tables-mcp-server
  - sagemaker-ai-mcp-server
  - stepfunctions-tool-mcp-server
  - well-architected-security-mcp-server
tools:
- name: aws___read_documentation
  description: Fetch full AWS doc pages as markdown.
  server: AWS Knowledge MCP Server
  input_properties: [requests]
- name: aws___search_documentation
  description: AWS docs search; each result's context is verbatim page text.
  server: AWS Knowledge MCP Server
  input_properties: [limit, search_phrase, topics]
- name: aws___list_regions
  description: Retrieve a list of all AWS regions.
  server: AWS Knowledge MCP Server
  input_properties: []
- name: aws___get_regional_availability
  description: AWS resource availability per region.
  server: AWS Knowledge MCP Server
  input_properties: [regions, resource_type, filters, next_token, region]
- name: aws___retrieve_skill
  description: Retrieve an AWS skill (workflows, references); returns SKILL.md.
  server: AWS Knowledge MCP Server
  input_properties: [file, skill_name]
oauth_discovery:
  probed: '2026-09-01'
  result: >-
    /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server on
    knowledge-mcp.global.api.aws both answer HTTP 200 with an HTML SPA shell, as does a negative-control
    path that cannot exist. No OAuth metadata is served; the endpoint is anonymous.