Give agents a multi-engine math kernel via MCP
An evidence-aware multi-engine mathematics kernel — usable both as a Python library (mathkernel) and as an MCP server (mathkernel-mcp) — so applications and LLMs can do advanced mathematics while preserving assumptions, provenance, and claim-specific evidence. Task: An evidence-aware multi-engine mathematics kernel — usable both as a Python library (mathkernel) and as an MCP server (mathkernel-mcp) — so applications and LLMs can do advanced mathematics while preserving assumptions, provenance, and claim-specific evidence. The LLM interprets intent; the MathKernel establishes mathematical evidence. Mathematical results carry an explicit trust level, an engine tag, and a derivation trail. Exact computation, checked certificates, symbolic results, certified enclosures, empirical evidence, and formal proofs are distinct claims. Exact arithmetic alone is not a formal proof; approximate-input ancestry must not silently disappear. The current development line is Beta. Optional engines and Studio have separate availability and validation boundaries; the package classification does not certify every backend, platform, or mathematical claim. Table of contents Why Architecture Feature matrix Installation Quickstart — MCP server Quickstart — Python library Trust model External formal-project audits Continuous symbolic mathematics Finite dynamics & PRNG analysis Engineering mathematics Geometry and topology Statistics and stochastic modeling PDEs and adaptive finite elements Relation and information-geometry inference Performance: numba · CUDA · parallelism Visualization & portable artifacts Shared multimodal projections Scientific sonification Unified multimodal artifacts MathKernel Studio MCP tool surface Configuration Repository layout Skill packages Testing Safety boundaries License Why LLMs are good at mathematical intent and bad at mathematical arithmetic. MathKernel inverts the division of labor: the model parses, plans, and interprets; the kernel computes and records claim-specific evidence. Some claims use independent certificates or cross-checks; others are exact computations in one engine. Engine agreement alone is not a proof, and a single trust label does not replace the evidence bundle. Architecture MathKernel is a typed orchestration layer rather than a single solver. The public facade owns parsing, contexts, object identity, persistence, evidence composition, resource policy and derivation tracking; domain adapters own the actual mathematics. Presentation layers sit downstream and cannot silently change the claim being made. This separation is deliberate: a renderer may present evidence, but it does not create stronger mathematical evidence merely by producing a polished plot or audio artifact. Feature matrix Typed functionality surface The generic MCP tools math_object_create, math_object_get, and math_apply expose the following compositional operations. This is the full typed-operation inventory; math_capability_query is the live source of parameter schemas, output types, limits, engines and verification methods.; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include Python environment setup and optional installation of engines like SymPy, Z3, or Lean. Known constraints involve non-product joint supports, branch-sensitive inputs, and transforms where SymPy cannot establish ROC. The package classification does not certify every backend or mathematical claim. Human approval is required for formal proofs, as engine agreement alone is not sufficient evidence..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
An evidence-aware multi-engine mathematics kernel — usable both as a Python library (mathkernel) and as an MCP server (mathkernel-mcp) — so applications and LLMs can do advanced mathematics while preserving assumptions, provenance, and claim-specific evidence.
Workflow
An evidence-aware multi-engine mathematics kernel — usable both as a Python library (mathkernel) and as an MCP server (mathkernel-mcp) — so applications and LLMs can do advanced mathematics while preserving assumptions, provenance, and claim-specific evidence. The LLM interprets intent; the MathKernel establishes mathematical evidence. Mathematical results carry an explicit trust level, an engine tag, and a derivation trail. Exact computation, checked certificates, symbolic results, certified enclosures, empirical evidence, and formal proofs are distinct claims. Exact arithmetic alone is not a formal proof; approximate-input ancestry must not silently disappear. The current development line is Beta. Optional engines and Studio have separate availability and validation boundaries; the package classification does not certify every backend, platform, or mathematical claim. Table of contents Why Architecture Feature matrix Installation Quickstart — MCP server Quickstart — Python library Trust model External formal-project audits Continuous symbolic mathematics Finite dynamics & PRNG analysis Engineering mathematics Geometry and topology Statistics and stochastic modeling PDEs and adaptive finite elements Relation and information-geometry inference Performance: numba · CUDA · parallelism Visualization & portable artifacts Shared multimodal projections Scientific sonification Unified multimodal artifacts MathKernel Studio MCP tool surface Configuration Repository layout Skill packages Testing Safety boundaries License Why LLMs are good at mathematical intent and bad at mathematical arithmetic. MathKernel inverts the division of labor: the model parses, plans, and interprets; the kernel computes and records claim-specific evidence. Some claims use independent certificates or cross-checks; others are exact computations in one engine. Engine agreement alone is not a proof, and a single trust label does not replace the evidence bundle. Architecture MathKernel is a typed orchestration layer rather than a single solver. The public facade owns parsing, contexts, object identity, persistence, evidence composition, resource policy and derivation tracking; domain adapters own the actual mathematics. Presentation layers sit downstream and cannot silently change the claim being made. This separation is deliberate: a renderer may present evidence, but it does not create stronger mathematical evidence merely by producing a polished plot or audio artifact. Feature matrix Typed functionality surface The generic MCP tools math_object_create, math_object_get, and math_apply expose the following compositional operations. This is the full typed-operation inventory; math_capability_query is the live source of parameter schemas, output types, limits, engines and verification methods.
Reported result
No measured business outcomes or performance metrics are reported in the source. The project is currently in Beta development. A proposed acceptance check is to verify that a sample calculation returns a valid derivation trail and correct trust label without human intervention, confirming the evidence-aware architecture functions as described.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include Python environment setup and optional installation of engines like SymPy, Z3, or Lean. Known constraints involve non-product joint supports, branch-sensitive inputs, and transforms where SymPy cannot establish ROC. The package classification does not certify every backend or mathematical claim. Human approval is required for formal proofs, as engine agreement alone is not sufficient evidence.
Try it
I need to set up MathKernel to solve a specific mathematical problem with full evidence tracking. First, install the library using 'pip install mathkernel[mcp]'. Then, use the MCP tool 'math_object_create' to define a [replaceable input: e.g., a symbolic integral or differential equation]. Apply the appropriate operation using 'math_apply'. Ensure the output includes the trust level, engine tag, and derivation trail. Before finalizing, I will manually review the evidence bundle to confirm the result is valid and the assumptions are preserved. If the trust level is not 'FORMAL' or 'EXACT', I will note the limitation. Do not proceed with any downstream actions until I confirm the mathematical evidence is sufficient for [replaceable context: e.g., engineering design or academic publication].
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.