generated: '2026-08-04'
method: searched
status: published-self-hosted
source: https://hammerspace.com/hammerspace-launches-ai-data-platform-based-on-nvidia-reference-design/
server:
  name: hammerspace
  vendor: Hammerspace Inc
  transport: unknown
  url: null
  hosted_by: customer
  description: 'Hammerspace ships Model Context Protocol services as a component of the Hammerspace AI
    Data Platform, announced at NVIDIA GTC in October 2025 and made generally available on an NVIDIA AI
    Data Platform reference design in March 2026. MCP sits alongside an integrated Milvus vector database:
    files tagged for embedding are promoted to Tier 0 (local NVMe on GPU servers), processed into vector
    embeddings, stored in Milvus for semantic search, and returned to their default tier — with MCP giving
    AI agents and RAG pipelines direct, policy-governed access to the resulting data estate. The server
    runs inside the customer''s own deployment of the platform, so there is no Hammerspace-hosted public
    MCP endpoint to introspect.'
discovery:
  public_endpoint: false
  tools_list_probe: not-applicable
  note: 'tools/list could not be probed: no public MCP URL is published, and every path on hammerspace.com
    is answered by a bot-protection challenge (HTTP 202). The tool set below is therefore NOT the official
    server''s manifest — it is recorded from the one public reference implementation, clearly attributed.'
tools: []
reference_implementation:
  name: MCP 1.5 — Advanced Hammerspace Storage Management with AI
  url: https://github.com/mbloomhammerspace/mcp-1.5-main
  license: MIT
  first_party: false
  attribution: Published on a personal GitHub account whose handle indicates Hammerspace affiliation,
    but which declares no organization and is not part of the github.com/hammer-space organization. Treated
    as a community/individual reference implementation, not a Hammerspace-published artifact.
  architecture: Drives Hammerspace through the hstk CLI against a mounted global file system and through
    the Anvil, with Kubernetes ingestion jobs and a Milvus vector database behind it.
  tools:
  - name: list_shares
    category: shares
  - name: list_files
    category: files
  - name: list_files_by_tag
    category: files
  - name: get_file_tags
    category: metadata
  - name: set_file_tag
    category: metadata
  - name: tag_directory_recursive
    category: metadata
  - name: create_objective
    category: objectives
  - name: apply_hs_objective
    category: objectives
  - name: remove_hs_objective
    category: objectives
  - name: apply_objective_to_path
    category: objectives
  - name: remove_objective_from_path
    category: objectives
  - name: list_objectives_for_path
    category: objectives
  - name: check_file_alignment
    category: placement
  - name: check_tagged_files_alignment
    category: placement
  - name: aiq_analyze_storage
    category: analysis
  - name: aiq_optimize_tiering
    category: analysis
  - name: list_jobs
    category: jobs
  - name: ingest_new_files
    category: ingestion
  - name: get_file_ingest_events
    category: ingestion
  - name: get_file_monitor_status
    category: ingestion
  - name: refresh_mounts
    category: mounts
related:
  vector_database: Milvus (integrated)
  tier0: Local NVMe on GPU servers pooled into the global namespace
  cli: cli/hammerspace-cli.yml
  rest_api: apis.yml#hammerspace:anvil-management-api
x-evidence:
  fetched: '2026-08-04'
  sources:
  - https://hammerspace.com/hammerspace-launches-ai-data-platform-based-on-nvidia-reference-design/
  - https://hammerspace.com/hammerspace-unveils-ai-data-platform-solution/
  - https://github.com/mbloomhammerspace/mcp-1.5-main
  caveat: Hammerspace's MCP support is asserted by Hammerspace in its own product announcements. Protocol
    version, transport, tool names and input schemas of the official server could not be verified because
    the server is deployed by the customer and no public endpoint or manifest is published.
deployment:
  mode: unclear
  verified: searched
  checked: '2026-08-12'
  source: catalog MCP census