Skip to content

v1.10.0: MCP

Latest

Choose a tag to compare

@xuzhougeng xuzhougeng released this 08 Sep 02:50

wisp-science v1.10.0

v1.10.0 brings scientific MCP tools into Wisp's native Rust runtime. The bundled catalog provides 23 connectors and 247 tools across roughly 80 upstream databases, with browsable connector introductions and tool parameters. This release also improves first-time setup, settings navigation, and Python/R skill execution.

Downloads

  • macOS (.dmg, signed + notarized): Apple Silicon (aarch64) and Intel (x86_64); compatible versions can update in-app.
  • Windows (MSI / NSIS): installers remain unsigned while SignPath release signing is disabled. Windows may show SmartScreen; review the publisher details before proceeding.
  • Linux (deb / AppImage): x86_64 and aarch64.

Installers are attached to Assets as the platform builds finish.

Highlights

Native Scientific MCP

  • Desktop, CLI, and the ACP bridge share the Rust wisp-bio catalog. Bundled scientific retrieval no longer launches the vendored Python MCP servers; Python is still available for analysis and custom MCP services that need it.
  • The catalog covers literature, omics archives, genes and ontologies, variants, proteins, structures, chemistry, clinical trials, and research resources. Sources include PubMed, GEO, UniProt, ChEMBL, OpenAlex, arXiv, bioRxiv/medRxiv, and ZINC.
  • Native coverage includes KEGG, CADD, PanglaoDB, and Sanger Cell Model Passports. Their academic-use, access, and snapshot conditions remain documented; upstream services still require network access and may impose limits.
  • Agents discover relevant tools before calling them. Full tool schemas are loaded as needed instead of being included in every turn.
  • The scientific MCP catalog documents the connectors, tool capabilities, upstream sources, and usage conditions.

Connections and Tool Details

  • Settings → Connections shows bilingual connector introductions, maintainers, and source/documentation links.
  • Expand a tool to inspect its description, parameter types, required fields, defaults, and full schema. Output schemas and examples appear when supplied by the service.
  • Per-tool allow/ask/deny controls remain accessible while browsing details. Project-wide approval modes now live in Settings → Permissions.
  • Plugin MCP processes on Windows receive the user-profile environment variables needed by their local dependencies.

Setup, Settings, and Skills

  • Local runtimes are optional during onboarding. Quick setup detects local tools and supports editing executable paths; keyboard navigation and error-card layout are more reliable.
  • Local environment and network settings are grouped under General. Separate proxy settings cover model APIs, MCP services, and local code; package-mirror guidance can be configured for environment installation.
  • Credentials use service detail pages, and Skills gain a catalog and package detail browser. Settings layouts keep controls reachable in smaller windows.
  • The app and documentation adopt SpicyChicken6's three-wisp icon and wordmark.

Python/R Runtime Reliability

  • Python script execution preserves script semantics, and runtime execution guidance is clearer.
  • Internal runtime generation guards stay out of model-facing tool schemas.
  • Skills can declare Python and R runtime sidecars through a consistent convention. Bundled helper paths, retrieval error handling, and loading side effects are corrected.
  • The public data acquisition skill includes optional geokit guidance.
  • The bundled models.dev catalog is refreshed for current model limits, capabilities, and pricing metadata.

Full Changes

All commits since v1.9.0:
v1.9.0...v1.10.0


wisp-science v1.10.0

v1.10.0 将科研 MCP 工具迁入 Wisp 的 Rust 原生运行时。内置目录提供 23 个连接器、247 个工具,覆盖约 80 个上游数据库,并支持直接浏览连接器介绍和工具参数。本次发布同时改进首次配置、设置导航,以及 Python/R 技能执行。

下载

  • macOS(.dmg,已签名 + 公证):提供 Apple Silicon(aarch64)和 Intel(x86_64)版本;兼容版本可在应用内更新。
  • Windows(MSI / NSIS):SignPath 正式签名仍处于停用状态,安装包暂未签名。Windows 可能显示 SmartScreen,请核对发布者信息后再继续。
  • Linux(deb / AppImage):提供 x86_64 和 aarch64 版本。

各平台构建完成后,安装包将陆续附加到下方 Assets。

重点功能

原生科研 MCP

  • 桌面端、CLI 与 ACP bridge 共用 Rust wisp-bio 目录。内置科研检索不再启动原先随应用打包的 Python MCP 服务;数据分析和需要 Python 的自定义 MCP 仍可使用 Python。
  • 目录覆盖文献、组学档案、基因与本体、变异、蛋白质、结构、化学、临床试验和科研资源,数据源包括 PubMed、GEO、UniProt、ChEMBL、OpenAlex、arXiv、bioRxiv/medRxiv 和 ZINC。
  • KEGG、CADD、PanglaoDB 和 Sanger Cell Model Passports 也已有原生实现。文档保留各数据源的学术用途、访问条件与快照说明;访问上游服务仍需网络,并受其使用限制约束。
  • Agent 先发现相关工具,再调用匹配能力;完整参数定义按需加载,不会在每一轮全部放入提示词。
  • 科研 MCP 目录 展示连接器、工具能力、上游来源与使用条件。

连接管理与工具详情

  • 设置 → 连接 展示连接器的中英文介绍、维护方,以及数据来源和官方文档链接。
  • 展开工具即可查看说明、参数类型、必填项、默认值及完整定义;服务提供返回数据定义或参数示例时,也会一并显示。
  • 浏览详情时可直接设置逐工具的允许、询问和禁止规则;项目级工具审批模式统一移至 设置 → 权限
  • Windows 插件 MCP 子进程可获得本地依赖需要的用户目录环境变量。

首次配置、设置与技能

  • 首次引导中的本地运行时改为可选;快速配置支持检测本地工具和编辑可执行文件路径,并改进键盘导航与错误卡片布局。
  • 本地环境与网络设置归入 常规。模型 API、MCP 服务和本地代码可分别配置代理,并可为环境安装设置包镜像指导。
  • 凭据按服务进入详情页;技能增加目录与包详情浏览。设置页面改进布局,让小窗口中的操作控件保持可达。
  • 应用与文档采用 SpicyChicken6 设计的三缕微光图标和文字标识。

Python/R 运行时可靠性

  • Python 脚本执行保留脚本语义,并明确运行时执行指导。
  • 内部运行时 generation 校验字段不再暴露到模型工具参数中。
  • 技能可通过统一约定声明 Python 和 R 运行时配套文件;修复内置辅助脚本路径、检索错误处理与加载副作用。
  • 公共数据获取技能增加可选的 geokit 使用指导。
  • 刷新内置 models.dev 目录,更新模型限制、能力与价格元数据。

完整变更

v1.9.0 以来的全部提交:
v1.9.0...v1.10.0

What's Changed

Full Changelog: v1.9.0...v1.10.0