Scicompute

io.github.CloudCorpRecords由 GitHub 账号 CloudCorpRecords 发布v0.1.0更新于 Nov 3, 2026

科学计算：命名张量、矩阵代数和向量运算，完全本地运行。

AI 翻译由 AI 模型根据下方的原文描述翻译，可能有误。

原文描述

Scientific computation: named tensors, matrix algebra and vector operations, fully local.

已验证STDIO仅桌面

概览

Scicompute 的本地 MCP 服务；无需登录；STDIO。

暂无 AI 概览：以下信息来自注册中心条目和服务本身。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Scicompute，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only，通过 STDIO。 STDIO 服务会启动本地进程，因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

scicompute

scicompute is a scientific-computation server built around one idea: a named, in-process tensor store. You create tensors under names, then run matrix algebra and vector operations by referring to those names. Everything is computed locally in pure TypeScript — no network, no external math library. One binary serves a command-line interface and a stdio MCP server.

This build is a labelled core subset of the documented tool set: tensor storage, linear algebra, and vector operations (14 tools). Eigen/SVD/QR decompositions, the vector-calculus tools, basis changes, and plotting from the wider documented set are not included; see the build report for the full omission list.

Install

bash
npx @reneturcios/scicompute create-tensor T 2,2 1,2,3,4

Or run the container image:

bash
docker run -i ghcr.io/cloudcorprecords/scicompute

The container defaults to the MCP server (see below). Pass CLI arguments to use it as a CLI instead.

CLI

bash
scicompute create-tensor A 2,2 4,7,2,6scicompute view-tensor Ascicompute determinant Ascicompute --version

The store is in-process, so a single CLI invocation holds its tensors only for its own lifetime; chained operations over several tensors belong to one MCP session, where the store persists across tool calls. Exit codes: 0 on success, 1 for a computation error (such as a shape mismatch or a singular matrix), 2 for usage errors.

Subcommands: create-tensor, view-tensor, delete-tensor, add, subtract, multiply, scale (--no-in-place to leave the stored matrix untouched), transpose, determinant, rank, inverse, dot, cross, project, and mcp.

MCP server

bash
npx -y @reneturcios/scicompute mcp

MCP client configuration:

json
{  "mcpServers": {    "scicompute": {      "command": "npx",      "args": ["-y", "@reneturcios/scicompute", "mcp"]    }  }}

Tools

ToolInputWhat it returns
create_tensorname, shape, values (all required)The stored tensor: name, shape, flat values, and nested data. values length must equal the shape product.
view_tensorname (required)The stored tensor under that name.
delete_tensorname (required)Confirmation that the tensor was deleted.
add_matricesa, b (required), result (optional)The element-wise sum matrix; stored under result when given.
subtract_matricesa, b (required), result (optional)The element-wise difference matrix.
multiply_matricesa, b (required), result (optional)The matrix product.
scale_matrixname, factor (required), in_place (optional)The scaled matrix. In place by default; stored matrix is untouched when in_place is false.
transposename (required), result (optional)The transposed matrix.
determinantname (required)The determinant of a square matrix.
rankname (required)The matrix rank, from row-echelon reduction.
matrix_inversename (required), result (optional)The inverse of a square, non-singular matrix.
vector_dot_producta, b (required)The dot product of two stored vectors.
vector_cross_producta, b (required)The cross product of two stored 3-vectors.
vector_projecta, b (required)The projection of vector a onto vector b.

Algebra and vector operands are tensor names, not raw data. Matrices are 2-entry shapes and vectors are 1-entry shapes; shape mismatches, missing tensors, non-square determinant/inverse requests, singular inverses, and projections onto the zero vector are all errors rather than silent zeros.

Development

bash
npm installnpm run build   # tsc -> dist/npm test        # node test runner over test/*.test.ts, no networknpm run dev     # run the CLI from source via tsx

License

MIT — see LICENSE.

来源：README.md，提交 0a9e447

工具

0
SourceWeft 尚未在沙箱中启动过该软件包，因此其工具未知。

版本历史

1
  1. v0.1.0最新Nov 3, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手；SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头，或保持端点能够响应探测，以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分