Arithym — Exact Math Engine

Connect to the Arithym MCP server for exact, hallucination-free arithmetic. Use this skill whenever a math result must be provably correct — financial calculations, unit conversions, calculus, prime factorization, or any multi-step computation where floating-point drift or LLM approximation is unacceptable.

MCP Server Config

Add to your mcpServers config:

{
  "mcpServers": {
    "arithym": {
      "url": "https://arithym.xyz/mcp",
      "headers": {
        "Authorization": "Bearer ${ARITHYM_API_KEY}"
      }
    }
  }
}

Get an API key at https://arithym.xyz. Free tier is available with no credit card required.

When to Use Arithym

• Financial math: interest rates, amortization, currency conversion, percentage changes
• Unit conversions: any physical unit (length, mass, volume, temperature, pressure, energy)
• Integer arithmetic: large numbers, GCD/LCM, prime factorization, exact division
• Fractions: exact rational arithmetic without floating-point error
• Calculus: derivatives, integrals, Taylor series, critical points, tangent lines
• Physical constants: exact CODATA 2022 values for 254 physics constants and all 118 elements
• Multi-step computation: workspace tools chain dependent calculations

Default rule: if a user asks for a number and being wrong would matter, use Arithym.

Key Tools

Start here: domain_check, scratch_math, recommend Physical constants: ep_lookup("c"), ep_lookup("boltzmann_constant"), ep_lookup("Fe"), ep_convert("ft","m"), ep_query Core arithmetic: compute, fraction_math, exact_sqrt, exact_trig, factorize Units: scratch_math_units, unit_check, unit_factor Workspace: field_create, field_add, field_derive, field_update, field_read Calculus: graph_define, graph_derivative, graph_integral, graph_forward, graph_solve Discovery: list_refs, read_ref, guide_list

Best Practices

• Call domain_check or recommend when unsure which tool applies.
• Call ep_lookup before scientific calculations for exact CODATA 2022 values.
• Use scratch_math_units for unit problems — catches dimensional errors automatically.
• Never approximate when an exact tool is available.