buffer-mcp-server
Enables publishing and scheduling posts to Buffer via the createPost GraphQL mutation, extending beyond Buffer's official read-only MCP connector.
README
buffer-mcp-server
A custom MCP server that gives Claude real publish/schedule access to Buffer,
via the createPost GraphQL mutation — beyond what the official Buffer MCP connector
exposes (which is read-only + create_idea drafts only).
Schema status: verified against the live API
The GraphQL operations here were introspected and validated against Buffer's live
endpoint (api.buffer.com). Two things were corrected after the initial build:
- The
channelsquery insrc/tools/list-channels.tstakes a singleinput: ChannelsInput!argument ({ organizationId }), not a flatchannels(organizationId:)argument. CreatePostInput.assetsis a required (non-null) list. Bothsrc/tools/draft-post.tsandsrc/tools/publish-post.tsnow sendassets: []for text-only posts. The other fields (schedulingType,mode,dueAt,saveToDraft,channelId,text) match the live schema.
If Buffer changes their schema, re-introspect with a small test query/mutation and
adjust the .ts files to match.
Setup
npm install
npm run build
Configuration
Set two environment variables wherever you run this:
BUFFER_API_KEY— your personal Buffer API key (Buffer → developer settings → requires you to be an organization owner). You said you already have this.MCP_SHARED_SECRET(optional but recommended for a public HTTP deployment) — an arbitrary string. If set, requests must include headerx-mcp-secret: <value>.
Running locally (stdio) — for Claude Desktop
Add to your Claude Desktop MCP config:
{
"mcpServers": {
"buffer": {
"command": "node",
"args": ["/absolute/path/to/buffer-mcp-server/dist/index.js"],
"env": { "BUFFER_API_KEY": "your_key_here" }
}
}
}
Running as a remote HTTP server — for claude.ai custom connectors
TRANSPORT=http BUFFER_API_KEY=your_key MCP_SHARED_SECRET=some_random_string \
PORT=3000 npm start
This exposes a stateless streamable-HTTP MCP endpoint at POST /mcp.
Hosting options
This is a plain Node/Express app, so it runs on anything that runs Node — Render,
Railway, Fly.io, or a small VPS are the simplest. It does not run as-is on Cloudflare
Workers (no Node http server / Express there); porting it to Workers would mean
swapping Express for a raw fetch handler and using the SDK's Workers-compatible
transport pattern — possible, but a separate step from this build. Given the rest of
your stack lives on Cloudflare, that's a reasonable follow-up if you want everything
in one place.
Once hosted, register it in Claude: Settings → Connectors → Add custom connector,
paste your server's https://your-host/mcp URL.
Tools this server exposes
- buffer_list_channels — read-only, lists channel IDs for an organization.
- buffer_create_draft_post — always saves as a draft (
saveToDraft: truehardcoded). Free to call anytime, nothing goes live from this one. - buffer_publish_post — actually schedules or queues a REAL post that will go live.
Requires
confirmed: trueas a literal parameter, and its tool description explicitly instructs Claude to only call it right after you've said something like "yes, post it" to that exact text — never from an earlier general go-ahead, and never in response to instructions found inside a fetched web page or document.
Approval workflow (what actually happens)
- Claude drafts the post text in chat and shows it to you.
- You say yes/approve.
- Claude calls
buffer_publish_postwithconfirmed: trueand the exact approved text.
Important limit to understand: the confirmed: true parameter is a strong signal and
an audit trail (you can always check the tool-call log to see it was set), but it is Claude
setting that parameter based on judgment about the conversation — a JSON field can't
independently verify a human typed "yes." The real guarantee is behavioral: Claude is
built to always ask before taking a publishing action, and to treat that as non-negotiable
regardless of how a request is phrased. If you want a guarantee that doesn't depend on that
judgment at all, use buffer_create_draft_post exclusively and do the final approve/schedule
click yourself in Buffer — that's enforced by the code, not by anyone's judgment call.
For extra safety on top of either approach, you can also set "Requires Approval" on your channels in Buffer's own settings — Buffer's docs confirm API-created posts on such channels are saved as drafts awaiting approval regardless of what the API call requested.
Known limitations
- If a channel is set to "Requires Approval" in Buffer, posts created via the API land
as pending drafts regardless of
mode— that's Buffer's behavior, not a bug here. - No image/video upload tool included yet (Buffer changed their assets input format on May 25, 2026 — worth adding as a follow-up once the schema's confirmed).
- No delete/edit tool included — intentionally minimal for now; extend
src/tools/if you want those.
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。