AI agents need multi-step scientific computing & statistical analysis. We give them one declaration.
Declarative computation graphs for AI tools. Compose tensor operations, statistical analyses, econometric tests, and portfolio optimisation in a single MCP call. Hardware-accelerated, vendor-agnostic backend.
Works with Claude · ChatGPT · Codex · Cursor · Zed
Add Euriklis to your AI agent
Run this command in your terminal.
claude mcp add euriklis npx @euriklis/mcp \
--env EURIKLIS_API_KEY=eur_free_...Need Codex or another agent? See the full install guide on /mcp.
Every operation your agent needs,
in one API call
Hardware-accelerated backend
Submit a computation graph and let the platform dispatch it to a vendor-agnostic, hardware-accelerated backend. No client-side setup, no runtime to install.
Declarative computation graphs
Describe a DAG of operations — tensors, matrix decompositions, statistical tests — in one JSON payload. No orchestration code.
MCP-native
First-class Model Context Protocol support. Claude, ChatGPT, Cursor, Zed and Codex can call the entire operation catalog through a single tool.
Linear algebra & tensor ops
LU, QR, Cholesky, eigenproblem, convolutions, pooling, norms, and 260+ matrix operations — all strictly validated before execution.
Statistics & econometrics
Box–Cox and Yeo–Johnson transforms, Gershgorin discs, and econometric tools callable from the same graph as the tensor operations.
Static validation
Every graph is checked for schema conformance, shape, arity, and cycles before any execution. Failed validation returns a structured diagnostic and consumes no compute allowance.
Built for agents doing real work
From prompt to matrix inverse
An agent receives a user's covariance matrix, normalises it, factors it with Cholesky, and returns the inverse — all in one tool call with no Python subprocess.
Tensor pipelines as data
Describe a forward pass — Conv2D, ReLU, pooling, softmax — as a JSON DAG. Euriklis validates the shapes and executes the pipeline on hardware-accelerated infrastructure.
Statistical tests on demand
An LLM agent runs Box-Cox normalisation, fits a regression, and tests residuals for heteroskedasticity — returned as structured JSON, no code interpreter needed.
Adjacency matrices at scale
Sparse graph operations, eigenvector centrality, and spectral decompositions — expressed as computation graphs and executed at native speed.
Ready to give your agent a math stack?
Free tier — 1,000 calls a month, no credit card. Install in Claude in 30 seconds.