ultrabrain-mcp
A local MCP reasoning gate that enables structured engineering thought through step-by-step reasoning, branching, merging, validation, and quality metrics without calling external LLM APIs.
README
ultrabrain-mcp
<p align="center"> <img src="./.github/assets/lcv-ideas-software-logo.svg" alt="LCV Ideas & Software" width="88" /> </p>
<p align="center"> <strong>LCV's local MCP reasoning gate for structured engineering thought.</strong> </p>
<p align="center"> <a href="https://www.npmjs.com/package/@lcv-ideas-software/ultrabrain-mcp"><img alt="npm" src="https://img.shields.io/npm/v/@lcv-ideas-software/ultrabrain-mcp.svg" /></a> <a href="https://github.com/LCV-Ideas-Software/ultrabrain-mcp/actions/workflows/ci.yml"><img alt="CI" src="https://github.com/LCV-Ideas-Software/ultrabrain-mcp/actions/workflows/ci.yml/badge.svg" /></a> <a href="https://ultrabrain-mcp.lcv.dev"><img alt="site" src="https://img.shields.io/badge/site-ultrabrain--mcp.lcv.dev-0f766e.svg" /></a> <a href="./LICENSE"><img alt="License" src="https://img.shields.io/badge/license-Apache--2.0-green.svg" /></a> </p>
Install. npm install -g @lcv-ideas-software/ultrabrain-mcp from npmjs.com, or npm install -g @lcv-ideas-software/ultrabrain-mcp --registry=https://npm.pkg.github.com from the GitHub Packages mirror.
Status. Latest release target: v01.01.00 for npm package 1.1.0. First publication started at v01.00.00. Public GitHub tags use the LCV display convention v00.00.00; npm keeps normal SemVer.
Release
| Release | Package | Date | Notes |
|---|---|---|---|
v01.01.00 |
1.1.0 |
2026-05-12 | Refinements for depth enforcement, active thought-reference validation, update record returns, real markdown merge output, optional file persistence, and expanded smoke coverage. |
v01.00.00 |
1.0.0 |
2026-05-12 | First LCV Ultrabrain MCP release: branded tool surface, session lifecycle, branching, merging, validation, analysis, prompts, resources, Pages, CI, publish, and StepSecurity baseline. |
What It Does
ultrabrain-mcp is an LCV-created MCP server for local, structured reasoning before engineering work is closed. It gives agents a single branded gate for:
- step-by-step reasoning with branches and revisions;
- quality metrics, confidence, budget mode, and meta checkpoints;
- bias detection and counterexample prompts;
- explicit evidence, assumptions, alternatives, risks, and next actions;
- session review, validation, metrics, export, prompts, and resources.
The server does not call external LLM APIs. It is a local MCP reasoning scratchpad and quality gate, so source code and private task context stay inside the active MCP host.
Tools
| Tool | Purpose |
|---|---|
ultrabrain_start |
Create a reasoning session and optionally seed the first thought. |
ultrabrain_think |
Append the canonical Ultrabrain reasoning step. |
ultrabrain_update |
Strengthen an existing thought with evidence, risks, confidence, or metrics. |
ultrabrain_branch |
Create or continue an alternative reasoning path. |
ultrabrain_merge |
Merge branch insights into a synthesis, best-evidence path, or decision. |
ultrabrain_validate |
Check evidence, alternatives, verification, bias, and closure gaps. |
ultrabrain_analyze |
Summarize quality, confidence, labels, bias counts, and recommendations. |
ultrabrain_review |
Render a session as summary, linear chain, tree, markdown, or JSON. |
ultrabrain_status |
Show session state, thought count, branches, and latest quality score. |
ultrabrain_history |
Return recent thought records. |
ultrabrain_export |
Export a session as markdown, text, or JSON. |
ultrabrain_metrics |
Return aggregate session and reasoning metrics. |
ultrabrain_templates |
List built-in LCV engineering reasoning templates. |
ultrabrain_reset |
Clear one session or all in-memory sessions. |
Canonical Example
{
"thought": "Map the failure, list evidence, compare alternatives, and choose the next verification step.",
"thought_number": 1,
"total_thoughts": 4,
"next_thought_needed": true,
"mode": "hybrid",
"step_type": "analysis",
"evidence": ["The failing behavior was reproduced locally."],
"alternatives": ["patch the caller", "patch the shared parser"],
"risks": ["a shared parser change can affect unrelated hosts"],
"next_actions": ["run the focused test", "inspect the shared call path"],
"quality_metrics": {
"logical_consistency": 4,
"completeness": 4,
"objectivity": 4,
"practicality": 5,
"clarity": 5
},
"response_format": "json"
}
MCP Configuration
Global npm installation target in the LCV Windows environment:
C:\npm-global\node_modules\@lcv-ideas-software\ultrabrain-mcp
Use the MCP server name ultrabrain in host configs:
{
"mcpServers": {
"ultrabrain": {
"command": "ultrabrain-mcp",
"args": []
}
}
}
Development checkout and global installation are intentionally separate. Do not create a junction or symlink from C:\npm-global\node_modules\@lcv-ideas-software\ultrabrain-mcp to the workspace checkout.
Optional local persistence can be enabled per host with ULTRABRAIN_STATE_DIR or ULTRABRAIN_PERSIST_DIR. When unset, sessions remain process-local.
Development
npm install
npm test
npm run format:public:check
npm pack --dry-run
The smoke test starts the built MCP server over stdio, lists tools, verifies the branded ultrabrain_* surface, exercises session start, thought, branch, merge, validation, status, export, prompts, and resources.
Release Automation
This repository follows the LCV package baseline:
- CI runs on
mainpull requests and pushes. - Dependabot tracks npm and GitHub Actions updates.
- Pages deploys the static site from
site/usingultrabrain-mcp.lcv.dev. - Auto-tagging derives padded public tags from
package.jsonversion. - Publish workflow releases to npmjs.com, GitHub Packages, and GitHub Releases.
- StepSecurity Harden-Runner is enabled on workflows.
Research
See docs/research-matrix.md. Ultrabrain imports ideas only at the architecture level. It does not copy third-party code, text, branding, non-English localization strings, or external model/API behavior.
Links
- Site: https://ultrabrain-mcp.lcv.dev
- npmjs.com: https://www.npmjs.com/package/@lcv-ideas-software/ultrabrain-mcp
- GitHub: https://github.com/LCV-Ideas-Software/ultrabrain-mcp
- Sponsors: https://github.com/sponsors/LCV-Ideas-Software
License
Apache-2.0. See LICENSE, NOTICE, and THIRDPARTY.
<p align="center"> Built by <a href="https://github.com/LCV-Ideas-Software">LCV Ideas & Software</a>. </p>
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。