matrix-compute-mcp
Server Details
P2P GPU matrix multiply, exact cosine search, and PCA/SVD verified via Freivalds' algorithm.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2024-11-05
- URL
- Repository
- matrixmultiplyconsortium/matrix-compute-mcp
- GitHub Stars
- 0
- Server Listing
- matrix-compute-mcp
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 5 tools
Each tool maps to a clearly distinct operation: similarity search, PCA/SVD, matmul execution, cost estimation, and network status. The only overlap is that estimate_job_cost and execute_distributed_matmul both concern matmul, but their actions are unambiguous.
All names use consistent snake_case and are descriptive. Most follow a verb-first pattern, though batch_similarity_search is adjective-first rather than verb-first.
Five tools is well-scoped for the server's purpose, covering core compute operations, cost estimation, and network discovery without unnecessary redundancy.
The core compute surface is covered, but there is no explicit job status, cancellation, or result-retrieval tool for long-running distributed jobs. Cost estimation also only supports matmul, leaving gaps for similarity-search and PCA/SVD jobs.
Available Tools
5 toolsbatch_similarity_searchCInspect
Runs exact cosine similarity search between query embeddings (M x D) and corpus embeddings (N x D) on distributed GPUs using OUTPUT_GRID_2D zero-reduction tiling.
Inputs: queries (required), corpus_embeddings (required), dtype, top_k, +1 more.
compute_pca_svdCInspect
Computes verified top-k Principal Component Analysis (PCA) / Singular Value Decomposition (SVD) on distributed GPU workers via randomized subspace iteration or stateful power iteration.
Inputs: k (required), matrix (required), center, method, +1 more.
estimate_job_costBInspect
Calculates exact upload/download network bandwidth (MB), task count, GFLOPs, and satoshi cost estimates for a distributed matrix multiplication C (m x n) = A (m x k) x B (k x n).
Inputs: k (required), m (required), n (required), dtype, +6 more.
execute_distributed_matmulBInspect
Executes verified distributed matrix multiplication C = A x B across decentralized GPU workers with dynamic Freivalds verification and L402 settlement. Accepts file paths (.npy, .npz, .safetensors) or inline 2D JSON arrays.
Inputs: matrix_a (required), matrix_b (required), dtype, tile_size, +4 more.
get_network_statusBInspect
Queries the Matrix Compute discovery relay for live GPU worker availability, total VRAM capacity, confidential computing (TEE) enclaves, and sats_per_gflop pricing.
Inputs: min_vram_mb, max_price_per_gflop.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
GPU fabric tools: NCCL benchmark analysis, degradation cost, cluster compatibility, XportL facts.
GPU-accelerated compute engine for BioTuring's biological databases
Run GPU acceptance tests on your NVIDIA GPU nodes over SSH: DCGM, GEMM, HBM/PCIe bandwidth, NCCL.
Measured AI-inference-storage benchmarks with citations, article search, KV-cache ROI estimation.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenancecuda4cpu runs CUDA and HIP programs on the CPU, unmodified, so AI agents can compile, run and test GPU kernels without a GPU. Its check finds the bugs a GPU hides at their source lines: missing __syncthreads(), races, out-of-bounds accesses, reads of uninitialized memory and leaks.6Apache 2.0
- AlicenseAqualityAmaintenanceeBPF-based GPU causal observability agent with MCP server. Traces CUDA Runtime and Driver APIs via kernel uprobes and host events via tracepoints to build causal chains explaining GPU latency. 7 tools: get_check, get_trace_stats, get_causal_chains, get_stacks, run_demo, get_test_report, run_sql. Telegraphic compression reduces token usage ~60%. Supports stdio and HTTPS (TLS 1.3) transport.1184-
- AlicenseNot gradedqualityNot gradedmaintenanceEnables agents to run zero-dependency statistical analysis natively over MCP, including Power Iteration PCA/SVD dimensionality reduction, Holt linear forecasting, modified Z-score and Tukey IQR anomaly detection, Welch's t-test, and multivariate gradient-descent regression on numerical data streams. Projects high-dimensional feature vectors into principal component representations without any external libraries.MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients to manage LLM inference memory through virtual paged-attention KV-cache block mapping, non-contiguous physical page allocation, zero-copy fragmentation tracking, radix-trie prefix caching, INT8 quantized compute, and speculative decoding verification. It also exposes prefill and decode latency telemetry so edge deployments can be benchmarked and tuned without external dependencies.7MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.