
Kontra Mcp
io.github.bluwork由 GitHub 账号 bluwork 发布v0.1.2更新于 Oct 24, 2026
The Tenth Man Protocol — structured contrarian analysis for decision-makers via MCP
概览
一个本地 MCP 服务器,返回结构化的反向分析提示,用四种模式对决策、计划或结论进行压力测试。
- 可以做什么
- 分析决策、计划或结论:返回一段对抗性提示,交由客户端自身的模型作答,可附带背景信息并选择模式(counter、probe、redteam、premortem)。
- 适用场景
- 适合在重大、难以逆转或已启动的决策前使用,例如架构选型、战略转向、发布、迁移、招聘计划,需要反面论证、尖锐提问、失败情景或事前复盘时。
- 运行要求
- 通过 stdio 在本地运行,通常用 npx 启动 kontra-mcp 这个 npm 包,因此需要 Node.js。无需 API 密钥或账号,使用 MCP 客户端已有的模型。可选环境变量 KONTRA_DEFAULT_MODE 和 KONTRA_LOG_LEVEL。
安装
在 SourceWeft 中
- 打开 控制台中的 Kontra Mcp,将其添加到工作区。
- 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
Kontra
The Tenth Man Protocol
"If nine agree, I disagree."
An MCP tool that stress-tests your decisions with structured contrarian analysis — so blind spots surface before consequences do.
What You Get
You say:
"Use kontra to challenge this: We should rewrite our backend from Node.js to Rust. Context: Current stack is Node.js, team of 3, shipping weekly."
Kontra returns:
BLIND SPOTS
-
You're measuring the cost of rewriting but not the cost of pausing. A team of 3 shipping weekly has momentum. The rewrite doesn't just cost engineering hours — it costs 3-6 months of feature development, customer feedback loops, and market positioning that your competitors won't pause for.
-
Rust's safety guarantees solve problems you may not have. If your Node.js backend isn't failing due to memory bugs or concurrency issues, you're optimizing for a class of problems that isn't hurting you. The rewrite is a solution — what's the problem it solves that your team is actually experiencing?
-
The team's Rust proficiency is unproven at production scale. Writing Rust that compiles is not the same as writing Rust that ships. Borrow checker fluency, async runtime decisions (Tokio vs. async-std), ecosystem maturity for your specific domain — these are all unknowns that compound under deadline pressure.
HIDDEN ASSUMPTIONS
...
THE STRONGEST COUNTER-ARGUMENT
...
WHAT-IF SCENARIOS
...
That's counter mode — the default. There are three more.
Quick Start
Claude Code
Then ask: "Use kontra to analyze my decision to..."
Claude Desktop
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
Cursor / VS Code
Add to your MCP settings:
Modes
counter — The Tenth Man
"Consensus is never to be mistaken for correctness."
The default. Assumes your decision is wrong and builds the strongest possible case against it.
Produces: Blind Spots, Hidden Assumptions, The Strongest Counter-Argument, What-If Scenarios
Best for: Major decisions where you need someone to steelman the opposing side. Architecture choices, strategy pivots, hiring plans, product bets.
probe — The Interrogator
"A good question does more damage to a bad decision than any argument ever could."
Asks 7-10 pointed questions across five lenses — no answers, no analysis, just the questions you haven't asked yourself.
Produces: Questions across Motivation, Evidence, Alternatives, Stakes, and Timing
Best for: Early-stage decisions where you suspect your reasoning has gaps but aren't sure where. Forces you to confront what you've skipped over.
redteam — The Red Team Commander
"Think like an adversary — not a hypothetical one, but a specific, intelligent opponent who wants this decision to fail."
Constructs 3-5 concrete failure scenarios with likelihood ratings, trigger events, cascade sequences, early warning signs, and mitigations.
Produces: Ranked failure scenarios with likelihood estimates, plus an Overall Vulnerability Assessment (1-10 scale)
Best for: Plans that are already in motion. Launch timelines, migration strategies, organizational changes — anything where you need to know exactly how it breaks.
premortem — The Postmortem From the Future
"Prospective hindsight increases the ability to identify reasons for future outcomes by 30%."
Imagines the decision was made 12 months ago and failed catastrophically. Writes the full postmortem.
Produces: Executive Summary, What Happened (Failure Narrative), Timeline, Root Causes, Signals We Missed, Lessons Learned
Best for: High-stakes, hard-to-reverse decisions. Useful when the team is confident and you need to break the spell of inevitability.
Usage
Counter a strategy decision:
"Use kontra to challenge this: We're going all-in on AI agents for our product roadmap."
Probe a career move:
"Use kontra in probe mode to question my decision to leave my job and start a SaaS company. Context: I have 8 months of runway, no co-founder, and a B2B idea with 3 LOIs."
Red team a launch plan:
"Use kontra in redteam mode against our plan to launch in 6 weeks. Context: We have no staging environment, 40% test coverage, and one DevOps engineer."
Pre-mortem a hiring decision:
"Use kontra in premortem mode on: We're hiring 5 junior engineers instead of 2 seniors. Context: Current team is 4 seniors, shipping velocity is the bottleneck."
Tool Parameters
What This Is NOT
- Not a pros-and-cons generator. Kontra doesn't weigh both sides. It attacks one side — yours — with full force.
- Not a risk assessment framework. No probability matrices or color-coded dashboards. Structured argumentation, not bureaucratic process.
- Not a generic "be careful" warning. Every output is specific to your decision and context. No boilerplate applies-to-anything advice.
- Not a replacement for human judgment. Kontra surfaces what you might miss. You still decide.
How It Works
Kontra uses the prompt-return pattern. When called, it constructs a carefully engineered adversarial prompt and returns it as the tool response. Your MCP client's own LLM processes the prompt and produces the analysis.
- No API keys needed — uses your client's existing LLM
- No additional network calls — prompt is returned, not executed
- Works with any MCP client — Claude Desktop, Claude Code, Cursor, VS Code
Analysis quality scales with model capability. Stronger models produce sharper, more specific contrarian analysis.
Configuration
License
Apache-2.0
来源:README.md,提交 c2399c2
工具
0版本历史
1- v0.1.2最新Sep 16, 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