Figma Context MCP
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
Generate AI video and images on Dare (Seedance, Veo, Kling, Hailuo), billed to your own Dare credits
选择你使用的 AI 应用,然后复制设置。
通过 3/3 项检查 · 在你的电脑上运行
第一步:你用哪个 AI?
选择你用的应用,步骤会显示在下方。
可连接 Claude,但需要先准备
不需要密钥,不用安装程序 · 仅限电脑上的 Claude 应用(Windows 或 Mac),网页和手机不可用
用 Dare 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude 说正在使用 Dare,或回答中出现 Dare 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
ChatGPT 暂时无法连接 Dare
Dare 需要安装在你的电脑上,而 ChatGPT 只能连接在网上运行的服务器
A · 使用能连接 Dare 的应用
B · 设计、图片与音频 类中能连接 ChatGPT 的其他服务器
从同类中使用较多的里挑选,仅限用地址即可连接、无需密钥的
Claude Code 暂时无法连接 Dare
开发者只提供了电脑版 Claude 应用的安装文件
A · 使用能连接 Dare 的应用
Cursor 暂时无法连接 Dare
开发者只提供了电脑版 Claude 应用的安装文件
A · 使用能连接 Dare 的应用
VS Code 暂时无法连接 Dare
开发者只提供了电脑版 Claude 应用的安装文件
A · 使用能连接 Dare 的应用
依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。
在你的电脑上运行,适用于
ClaudeGooseLM StudioMsty StudioCherry StudioVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月14日
注册名称已通过 GitHub 账户 avi-aggarwal14 验证,该账户拥有此仓库
代码最后更新于20天前
使用 MIT
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
提问、分享技巧或报告无法使用。按“有帮助”数量排序。
还没有评价
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 3 项适用的基础检查中通过了 3 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于20天前,最新版本(0.2.0)于29天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 MCPB。
同一分类中有:Figma Context MCP, AutoClip, Next AI Draw.io
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
draw.io diagrams your AI agent draws on a live browser canvas that you can also edit by hand
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
Generate video and images from inside Claude, billed to your own [Dare](https://trydare.com) credits.
Ask Claude for a shot and it makes one — Seedance 2.5, Seedance 2.0, Veo 3.1, Kling 3.0 Pro, Hailuo 2.3 for video, Nano Banana 2, GPT Image 2 and Seedream 5 Pro for stills. No new subscription, no API key to buy: it spends the Dare credits you already have.
> make me a 10 second 720p cinematic shot of rain on a neon Tokyo street, 16:9
Estimated 110 credits (you have 4,280). Generating… Done: https://storage.trydare.com/…/rain-tokyo.mp4
Works with Claude Code, the Claude desktop app, and Cowork.
[!IMPORTANT]
Dare does not publish a developer API. This server talks to the same private endpoint the
Dare web app uses, signed in as you with your own browser session. It works today — every
tool here has been exercised against Dare's live servers, including a generation run to
completion — but Dare can change that contract without notice, and using it may sit outside
Dare's terms of service. Your account, your credits, your call.
You need [Node.js](https://nodejs.org) 20 or newer. Check with node --version; if that errors, install the LTS build from nodejs.org (on a Mac, brew install node also works).
Then run one command in your terminal:
npx dare-mcp-server setup
That is the whole install. The wizard:
~/.dare-mcp/config.json with owner-only permissions,Restart Claude when it finishes. In Claude Code, /mcp should now list dare.
npx dare-mcp setup works too — dare-mcp is a short alias that forwards to the same package.
Claude desktop app, no terminal: download dare-mcp-server-0.2.0.mcpb from the mcpb-0.2.0 release, open it, and paste your Dare token when asked. It is also listed on the official MCP Registry as io.github.avi-aggarwal14/dare-mcp-server.
Step 2 in detail: finding your Dare token
The server signs in as you, using the long-lived Clerk __client cookie from your browser. It mints a fresh 60-second session token for every call, exactly as the Dare web app does.
https://clerk.trydare.com. (In Firefox and Safari the tab is called Storage instead.)__client and copy its Value — a long string starting eyJ. Copy the value, not the name.The cookie is valid for about a year, so this is a one-time step unless you sign out of Dare.
Treat it like a password. It grants full access to your Dare account and can spend your credits. The wizard keeps it in one file with 0600 permissions and deliberately does *not* write it into your Claude config, so your Claude settings stay safe to sync or share.
Prefer to do it by hand?
Claude Code
claude mcp add dare --scope user -- npx -y dare-mcp-server
Claude desktop app — edit the config file:
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.json{
"mcpServers": { "dare": { "command": "npx", "args": ["-y", "dare-mcp-server"] } } }
Then quit and reopen the app.
Both of these read your token from ~/.dare-mcp/config.json. If you would rather pass it as an environment variable (CI, containers, a shared machine), set DARE_CLIENT_TOKEN instead — it takes precedence over the stored file. See Configuration.
Running from source
git clone https://github.com/avi-aggarwal14/dare-mcp-server.git
cd dare-mcp-server npm install # builds automatically npm run setup # same wizard
Point Claude at node /absolute/path/to/dare-mcp-server/dist/cli.js instead of npx.
npx dare-mcp-server check
Mints a token, reads your balance, lists a few library items. Run it any time the tools start failing — it is the fastest way to tell a stale cookie from a Dare outage.
Once Claude has restarted, just ask in plain English:
> how many Dare credits do I have?
what video models can I use, and what do they cost?
make a 5 second clip of a paper boat going down a rain gutter, cinematic, 16:9
use ~/shots/hero.png as a reference and animate the character walking toward camera
what's in my Dare library from this week?
Claude will quote you a credit estimate before spending anything, and you can always ask it to do a dry run first.
| Tool | Spends credits | What it does |
|---|---|---|
dare_list_models | no | Models with durations, qualities, aspect ratios, reference limits |
dare_estimate_cost | no | Local credit estimate before you commit |
dare_get_credit_balance | no | Current balance |
dare_generate_video | yes | Text-to-video, Seedance 2.5 by default; optional reference media. dry_run: true previews the spec and price for free |
dare_generate_image | yes | Stills via Nano Banana 2, GPT Image 2, Seedream 5 Pro. Supports dry_run |
dare_get_generation | no | Poll status, get the output URL |
dare_list_generations | no | Paginated library listing |
dare_cancel_generation | no | Stop an in-flight job |
dare_delete_generation | no | Delete from the library (destructive) |
dare_upload_media | no | Upload a reference image/clip/audio, returns a storage_key |
dare_list_uploads | no | Previously uploaded references |
dare_delete_upload | no | Delete an uploaded reference (destructive) |
dare_list_projects | no | Your Dare projects |
| Command | What it does |
|---|---|
npx dare-mcp-server setup | Interactive first-run wizard |
npx dare-mcp-server check | Verify credentials, print credit balance |
npx dare-mcp-server | Start the MCP server on stdio (Claude runs this for you) |
npx dare-mcp-server http | Optional HTTP transport for self-hosting |
npx dare-mcp-server --help | Everything above |
Credentials are read in this order, first hit wins:
DARE_CLIENT_TOKEN in the environment~/.dare-mcp/config.json (written by setup)The stored file is plain JSON, so you can edit it directly:
{
"clientToken": "eyJ...", "maxCreditsPerCall": 500, "uploadRoots": ["/Users/you/Movies", "/Users/you/Pictures"] }
Everything else is environment variables:
| Variable | Default | Purpose |
|---|---|---|
DARE_CLIENT_TOKEN | from stored config | The __client cookie from clerk.trydare.com. Required one way or the other. |
DARE_CONFIG_DIR | ~/.dare-mcp | Where the stored config lives. |
DARE_MAX_CREDITS_PER_CALL | 500 | Refuse any single generation estimated above this many credits. 0 disables. |
DARE_UPLOAD_ROOTS | — (blocked) | Comma-separated directories dare_upload_media may read local files from. Empty means local file reads are refused. |
DARE_ALLOW_URL_UPLOADS | true | Allow uploading from a public URL. Private and link-local addresses are always refused. |
DARE_MAX_UPLOAD_BYTES | 536870912 | Largest upload accepted. |
DARE_SESSION_TOKEN | — | A ready-made 60s session JWT. Testing only. |
DARE_SESSION_ID | auto | Pin a specific sess_... id instead of auto-detecting. |
DARE_REQUEST_TIMEOUT_MS | 120000 | Per-request timeout. |
DARE_SERVER_URL | https://api.trydare.com | Dare RPC base. |
DARE_CLERK_FAPI_URL | https://clerk.trydare.com | Clerk Frontend API base. |
DARE_MAX_CREDITS_PER_CALL is a client-side circuit breaker against a model talking itself into a 10-variation 30-second batch (3,200 credits). The default of 500 allows any single Seedance 2.5 clip while blocking runaway batches; raise it deliberately when you want a batch. The estimate is conservative in the one place it cannot be exact: a video reference whose length Dare cannot report is priced at 30 seconds, which is Dare's own assumption. The guard fails closed: if a reference clip's duration cannot be read, the generation is refused rather than submitted on a guessed estimate.
DARE_UPLOAD_ROOTS is an allowlist, not a convenience. Without it, dare_upload_media refuses local paths entirely — otherwise a prompt-injected model could ask the server to upload your shell environment or Claude config, both of which contain DARE_CLIENT_TOKEN. Point it at the folders that actually hold your source footage.
Dare prices every job as provider cost in dollars, times 1.5, converted at 750 credits per $49.99, rounded up to the nearest 1, 5 or 10 credits. The server reproduces that formula exactly; it was checked against real charges (a 4s 480p Seedance 2.5 clip billed 20 credits, as predicted). Typical figures:
| Job | Credits |
|---|---|
| Seedance 2.5, 4s, 480p | 20 |
| Seedance 2.5, 8s, 720p | 90 |
| Seedance 2.5, 10s, 720p | 110 |
| Seedance 2.5, 30s, 480p | 150 |
| Seedance 2.5, 30s, 720p | 320 |
| Seedance 2.0 Fast, 5s, 480p | 15 |
| Seedance 2.0, 15s, 4K | 530 |
| Veo 3.1, 8s, 1080p | 80 |
| Kling 3.0 Pro, 5s | 20 |
| Hailuo 2.3 Pro (fixed 6s) | 15 |
| Nano Banana 2 image, 2K | 4 |
| GPT Image 2, 1:1 | 5 |
Attaching a video reference applies a 0.6x multiplier on Seedance models, and on Seedance 2.5 the reference clip's own seconds are billed on top. Image and audio references do not change the price.
Always sanity-check with dare_estimate_cost before a batch.
1. dare_upload_media { file_path: "/Users/you/shots/hero.png" }
-> storage_key: "uploads/abc123..."
Seedance 2.5 takes up to 50 references (30 image, 10 video, 10 audio), each clip 2–30s, 30 combined seconds per kind. With a video reference attached it derives duration and aspect ratio automatically. Seedance 2.0 models take 12 (9/3/3), 15s combined, and need a visual reference alongside any audio one. Veo 3.1 takes up to 3 images and is then fixed at 8 seconds. Kling and Hailuo take no references at all.
A stateless streamable-HTTP entrypoint is included for hosting the server behind a URL:
DARE_CLIENT_TOKEN='eyJ...' \
DARE_MCP_BEARER_TOKEN="$(openssl rand -hex 32)" \ PORT=3000 npx dare-mcp-server http
POST /mcp, health check at /health. The server refuses to start without a bearer token of at least 24 characters, since these tools spend real money. DNS-rebinding protection is on by default and scoped to localhost; widen it with DARE_ALLOWED_HOSTS / DARE_ALLOWED_ORIGINS w