Stet MCP Server
Enables drafting, sending, and measuring Slack broadcasts from any MCP-capable agent using the Stet platform.
README
@stethq/mcp
Local stdio MCP server for Stet. Draft, send, and measure Slack broadcasts from any MCP-capable agent.
Stet is a Slack-first broadcasting tool: compose a message once, target it per audience, preview the real Slack render, then send or schedule it across many channels with delivery tracking. This package exposes that whole workflow as MCP tools, so an agent can draft a broadcast, preview exactly what will land, send it, and read back who replied.
You need a Stet workspace to use this. Sign up at stethq.com.
Install
Nothing to install — npx runs the server on demand.
claude mcp add stet \
--env STET_API_KEY=bk_xxxxxxxxxxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx \
--env STET_API_URL=https://app.stethq.com \
-- npx -y @stethq/mcp
Or add it to your MCP config by hand:
{
"mcpServers": {
"stet": {
"command": "npx",
"args": ["-y", "@stethq/mcp"],
"env": {
"STET_API_KEY": "bk_xxxxxxxxxxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
"STET_API_URL": "https://app.stethq.com"
}
}
}
}
Generate an API key at Dashboard → Settings → API keys. Keys are workspace-scoped; pick the write scope for full functionality (read-only keys can list and measure, but cannot compose or send).
Requires the Growth plan or higher. The REST/MCP surface is gated: below
Growth, every call returns 403 with a plan-upgrade-required problem
document. If you do not have a workspace yet, start at
stethq.com — mention that you want API/MCP access and we
will get you on a plan that includes it.
Typical flow
list_sending_profiles → pick who it comes from
list_channels → find the channels to hit
create_broadcast → returns a draft id
set_one_time_channels → target ad-hoc channels (or pass audience_ids up front)
set_broadcast_content → write the whole body as Markdown
preview_broadcast → one audience_id at a time (from list_audiences or
the set_one_time_channels response) — see the exact
Slack Block Kit that will land
send_broadcast_now → or schedule_broadcast
—— later ——
get_broadcast_sends → who got it, who failed, and why
get_broadcast_reactions → emoji engagement, by channel
get_broadcast_replies → what people actually said back
Tools
Compose
| Tool | Description |
|---|---|
create_broadcast |
Create a draft with a title, sending profile, and optional target audiences. Returns the new id. |
set_broadcast_content |
Compose (or replace) the entire body from Markdown — one string, or per-audience segments. |
add_block |
Append a single content block (section, header, divider, image, context, actions, rich_text_list, fields). Prefer set_broadcast_content for whole bodies. |
set_block_visibility |
Restrict a block to a subset of the target audiences. Empty array = visible to all. |
reorder_blocks |
Reorder blocks by supplying the full block-id list in the new order. |
Target
| Tool | Description |
|---|---|
list_channels |
List Slack channels in the connected workspace. Free-text query plus structural filters. |
list_audiences |
List saved audiences (named sets of channels). |
create_audience |
Create a saved audience from an explicit list of channel ids. |
set_one_time_channels |
Target a draft at an ad-hoc channel set without creating a saved audience. Empty list deletes the group. |
suggest_audiences |
Heuristic audience suggestions based on channel-name patterns. |
list_sending_profiles |
List sender identities configured in this workspace. |
create_sending_profile |
Create a sender identity: display name, avatar, footer, default audiences. |
Check and send
| Tool | Description |
|---|---|
validate_broadcast |
Check for channel-overlap conflicts (one channel in several audiences whose visible blocks differ). |
preview_broadcast |
Render the per-audience Slack Block Kit exactly as it will be delivered. |
send_broadcast_now |
Send immediately. If the broadcast is already scheduled, returns 409 with the pending send time — retry with confirm: true to discard that schedule and send now. |
schedule_broadcast |
Schedule for an ISO timestamp + timezone. |
cancel_broadcast |
Cancel a scheduled or in-flight broadcast. Already-posted messages stay posted. |
Measure
| Tool | Description |
|---|---|
list_broadcasts |
List broadcasts newest-first with status, counts, send time, and a content preview. Optional status filter. |
get_broadcast_status |
A broadcast's content and lifecycle state: row, target audiences, blocks. |
get_broadcast_sends |
Per-channel delivery: status, error message, Slack permalink, reply/reaction counts, plus a delivered/failed/pending summary. |
get_broadcast_reactions |
Emoji reaction totals — overall, by emoji, and by channel. |
get_broadcast_replies |
In-thread replies with the replier's display name and a permalink. Optional per-channel filter. |
Reactions and replies are only captured for messages Stet itself posted — they are matched to a send by channel + timestamp. Messages posted to Slack by other means have no engagement data here.
Engagement history is limited to your plan's retention window: 30 days on Growth and Scale, unlimited on Enterprise. Rows outside the window are hidden, not deleted, and reappear if you upgrade.
Changing the tool schemas
The schemas in src/tools.ts describe a REST API that lives in a separate,
private repository — there's no compiler tying the two together, so drift is
possible.
Be extra careful with output schemas. Validation on the output side is deliberately fail-soft: a wrong output schema still returns usable data to the caller and only logs a warning to stderr. A mistake here does not show up as a visible failure to anyone using the server.
This repo's unit tests run against a mock and can't catch that kind of drift —
the mock fabricates responses that match whatever the schema says, so it can't
tell you the schema itself is wrong. The only check that verifies these schemas
against the real API is a conformance suite that lives in the API repo
(npm run mcp:verify). Pair any release that touches src/tools.ts with a run
of it.
Environment
| Variable | Required | Description |
|---|---|---|
STET_API_KEY |
Yes | Workspace-scoped API key (Dashboard → Settings → API keys). |
STET_API_URL |
Yes | Base URL of your Stet deployment, e.g. https://app.stethq.com. |
License
MIT — see LICENSE.
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。