Compute Resources MCP Server

MCP server for ACCESS-CI compute resources including hardware specifications, capabilities, and configurations.

Data source and its limits

This server reads CiDeR, the ACCESS resource catalog, via the operations API. Resource providers are responsible for keeping their CiDeR records current and largely do not, so treat what it returns as incomplete rather than authoritative. Verified 2026-09-20:

• get_resource_hardware reports Delta at 124 CPU nodes where it has 132, omits the MI210 in Delta's AMD node, and lists Bridges-2 as H100-only when it also has L40S and V100. NCSA's and PSC's own user guides agree with the curated documentation, not with CiDeR.
• It returns no hardware specifications at all for Stampede3, KyRIC, Voyager or Neocortex — the description is prose with no node counts or memory.
• has_gpu includes resources whose hardware payload has no GPU block and excludes non-GPU accelerators such as Habana Gaudi.
• A resource's display name may carry a hand-typed COMING SOON or NO NEW ALLOCATIONS label that contradicts its own accessAllocated flag.

For hardware specifications, prefer the curated per-resource pages under https://support.access-ci.org/documentation/resources, which are maintained against the resource providers' own documentation. Use this server to enumerate and filter the catalog, and to resolve resource IDs for other ACCESS services.

Usage Examples

Discovery & Search

"List all GPU resources"
"Resources at NCSA"
"Cloud computing systems"
"Delta hardware specifications"

Recommendations

"Recommend a resource for machine learning with GPUs"
"What system should I use for molecular dynamics?"
"Best resource for a beginner doing genomics analysis"

Tools

search_resources

Search and filter ACCESS-CI compute resources. Returns resource IDs usable by other ACCESS-CI services.

Parameters:

Parameter	Type	Description
id	string	Get specific resource (e.g., "delta.ncsa.access-ci.org")
query	string	Search names, descriptions, organizations
type	enum	Filter: compute, storage, cloud, gpu, cpu
has_gpu	boolean	Filter for GPU resources
organization	string	Filter by org: NCSA, PSC, Purdue, SDSC, TACC
limit	number	Max results (default: 50)

Examples:

// List all resources
search_resources({})

// Get specific resource
search_resources({ id: "expanse.sdsc.access-ci.org" })

// Find GPU resources at NCSA
search_resources({ has_gpu: true, organization: "NCSA" })

get_resource_hardware

Get detailed hardware specs (CPU, GPU, memory, storage) for a resource.

Parameters:

Parameter	Type	Description
id	string	Resource ID (required)

Example:

get_resource_hardware({ id: "delta.ncsa.access-ci.org" })

Prompts

recommend_compute_resource

Get personalized resource recommendations based on research needs.

Arguments:

Argument	Required	Description
research_area	Yes	Field of research (e.g., "machine learning", "molecular dynamics")
compute_needs	Yes	Requirements (e.g., "GPU for training transformers", "high memory for genome assembly")
experience_level	No	HPC experience: beginner, intermediate, advanced
allocation_size	No	Scale needed (e.g., "small pilot project", "large-scale production")

Installation

npm install -g @access-mcp/compute-resources

Configuration

{
  "mcpServers": {
    "access-compute-resources": {
      "command": "npx",
      "args": ["@access-mcp/compute-resources"]
    }
  }
}

Resources

• accessci://compute-resources - All compute resources
• accessci://compute-resources/capabilities-matrix - Resource comparison matrix
• accessci://compute-resources/gpu-guide - GPU selection guide
• accessci://compute-resources/resource-types - Resource type taxonomy

---

Package: @access-mcp/compute-resourcesVersion: v0.9.0 Main: dist/index.js