M
MethodicDev/candor
Honest agent UI for shadcn/ui: chat, tool calls and permission prompts that never overstate what an agent did
在 GitHub 查看 ↗链接已于 2026-11-12 核查 · HTTP 200 · 打开的是厂商自己的官网
安装与配置
客户端配置（取自该仓库的安装命令）
{
"mcpServers": {
"methodicdev-candor": {
"command": "npx",
"args": ["degit"]
}
}
}解析自我们于 2026-11-09 抓取的该仓库 README；我们未执行过。
command/args 来自仓库写明的安装命令；不确定时请以该项目 README 为准。
VS Code（.vscode/mcp.json —— 键名是 servers）
{
"servers": {
"methodicdev-candor": {
"command": "npx",
"args": ["degit"]
}
}
}VS Code 顶层键是 <code>servers</code>，其余与上面那份相同。
Claude Code（命令行）
claude mcp add --transport stdio methodicdev-candor -- npx degit
Claude Code 不收 JSON 文件，用这条命令直接加。
配置项的键名以你的客户端为准：客户端与 CLI
- ★ 数取值
- 0 ★ · MethodicDev/candor · 取自 GitHub 搜索 API · （2026-11-13 20:00:50 核对）
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型：开源项目
- 收录与活跃
- 收录于 2026-11-10 · 最近见到 2026-11-14 · 共见到 5 次 · 被 1 个来源列出
- 评分构成
- 综合分 103 · 1 个来源 · repo
- 安装
npx degit
常见问题（答案都来自我们的记录）
怎么安装？
该仓库给出的安装命令是
npx degit。我们的记录里它是仓库自己提供的；不保证在你的客户端上直接可用。它适合谁 / 用来做什么？
我们把它归在这些标签下: 开源项目。标签来自条目自身，不是我们强加的分类。
谁维护它、还在更新吗？
仓库
MethodicDev/candor；我们收录于 2026-11-10；最近见到 2026-11-14。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。# Candor UI: the free tier
以下摘录自该仓库的 README（片段） · 在 GitHub 读完整 README →
Honest agent UI for shadcn/ui: chat, tool calls and permission prompts that never overstate what an agent did. MIT.
Try the app demo · Honest against typical, side by side · Docs · Changelog
Interface components for agent apps: a thread, messages, a composer, tool calls, reasoning and citations, the
building blocks under them, a light, dark and high-contrast theme, and adapters for the Vercel AI SDK, the Claude
Agent SDK and AG-UI. For shadcn/ui projects, installed with the candor-ui CLI or shadcn's own, MIT licensed.
This repo is the source of Candor UI's 69 free items (version 2.6.2), the skill and plugins for coding
agents and three starter apps. Docs and live examples: https://candor.themethodic.com.
The same free items are on npm as @candor-ui/react, for people who
would rather npm install than copy code in. That package changes only when a free item changes, so its version can be
behind the kit's after a release that changed only Pro items; the CLI, candor-ui, has a version of its own. The
changelog says which release changed what.
## Try it in a minute, no key
bash
npx degit MethodicDev/candor/starter-vite-ag-ui my-app
cd my-app
npm install
npm run dev
Open the address it prints. A scripted agent answers in the page: fixed replies, no model, no server and no API key,
and the page says so. Ask it to add "Try a Candor block" to your list and you'll see a line of its reasoning, a tool
call marked "Needs you", and a permission prompt that shows what it will run, with Deny and Allow and neither
chosen for you. That is what the image above shows. Point VITE_AGENT_URL at an AG-UI agent (LangGraph, CrewAI…)
when you want a real one.
## Why Candor
An agent that acts on real systems can mislead without saying anything false: a spinner nobody has heard from, a total
that treats a missing cost as zero, an approval that now applies to a different change. Candor's components keep these
9 rules. Where Honest against typical has a moment that shows one, the link goes to its
Typical and Honest switch; the others go to Principles.
Model output is text, never markup: Adapters read what a model writes as a small safe subset of markdown; a link becomes plain text with its address beside it.
Unknown stays unknown: A cost or duration the service didn't report reads “not measured”, not zero, and an estimate says it is one.
A control that can't be used says why: A control that can't be used carries its reason, on hover and focus and to screen readers.
Nothing is pre-selected in a decision: Approve and reject, allow and deny start with neither chosen.
Approve the exact change: A decision is tied to a hash of the change that was asked about; a change that moved since is refused.
Reasoning is believed, not verified: Reasoning is shown as the model's own account and marked as unchecked; evidence is shown as evidence.
The agent is always named, by the app: Every agent message names its agent from the app's data, never from model output.
Say what can and can't be undone: Destructive actions say what they remove and whether it can be brought back.
Motion follows the outcome: A success mark is drawn after the result settles; motion reports a change and never predicts one.
Try the app demo · Honest against typical, side by side · Docs · Changelog
Interface components for agent apps: a thread, messages, a composer, tool calls, reasoning and citations, the
building blocks under them, a light, dark and high-contrast theme, and adapters for the Vercel AI SDK, the Claude
Agent SDK and AG-UI. For shadcn/ui projects, installed with the candor-ui CLI or shadcn's own, MIT licensed.
This repo is the source of Candor UI's 69 free items (version 2.6.2), the skill and plugins for coding
agents and three starter apps. Docs and live examples: https://candor.themethodic.com.
The same free items are on npm as @candor-ui/react, for people who
would rather npm install than copy code in. That package changes only when a free item changes, so its version can be
behind the kit's after a release that changed only Pro items; the CLI, candor-ui, has a version of its own. The
changelog says which release changed what.
## Try it in a minute, no key
bash
npx degit MethodicDev/candor/starter-vite-ag-ui my-app
cd my-app
npm install
npm run dev
Open the address it prints. A scripted agent answers in the page: fixed replies, no model, no server and no API key,
and the page says so. Ask it to add "Try a Candor block" to your list and you'll see a line of its reasoning, a tool
call marked "Needs you", and a permission prompt that shows what it will run, with Deny and Allow and neither
chosen for you. That is what the image above shows. Point VITE_AGENT_URL at an AG-UI agent (LangGraph, CrewAI…)
when you want a real one.
## Why Candor
An agent that acts on real systems can mislead without saying anything false: a spinner nobody has heard from, a total
that treats a missing cost as zero, an approval that now applies to a different change. Candor's components keep these
9 rules. Where Honest against typical has a moment that shows one, the link goes to its
Typical and Honest switch; the others go to Principles.
Model output is text, never markup: Adapters read what a model writes as a small safe subset of markdown; a link becomes plain text with its address beside it.
Unknown stays unknown: A cost or duration the service didn't report reads “not measured”, not zero, and an estimate says it is one.
A control that can't be used says why: A control that can't be used carries its reason, on hover and focus and to screen readers.
Nothing is pre-selected in a decision: Approve and reject, allow and deny start with neither chosen.
Approve the exact change: A decision is tied to a hash of the change that was asked about; a change that moved since is refused.
Reasoning is believed, not verified: Reasoning is shown as the model's own account and marked as unchecked; evidence is shown as evidence.
The agent is always named, by the app: Every agent message names its agent from the app's data, never from model output.
Say what can and can't be undone: Destructive actions say what they remove and whether it can be brought back.
Motion follows the outcome: A success mark is drawn after the result settles; motion reports a change and never predicts one.
收录徽章（可选）
想把收录标记放进 README 就复制这行，可选，不影响收录。
[](https://anclum.com/zh/s/methodicdev-candor/)
- 来源
- github · https://github.com/MethodicDev/candor
- 按领域浏览
- 开源项目
- 收录于
- 今天
同类推荐
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。