MCP

Market alerts your AI assistant can set up

The Subtrigger MCP server gives AI assistants 25 tools to find markets, read live values and history, and create, change and pause alerts, all with your API key.

Subtrigger has a remote MCP server (Model Context Protocol, Streamable HTTP) at https://subtrigger.com/mcp. With an API key, Claude Code, Cursor, VS Code or your own agent can search crypto and stock markets, read live values and alert history, and create, change, pause and delete alerts in plain conversation.

Connect in one command

Claude Code:

claude mcp add --transport http subtrigger https://subtrigger.com/mcp \
  --header "Authorization: Bearer st_…"

Cursor, VS Code and other clients that take a JSON config:

{
  "mcpServers": {
    "subtrigger": {
      "url": "https://subtrigger.com/mcp",
      "headers": { "Authorization": "Bearer st_…" }
    }
  }
}

The app shows both snippets with your key filled in right after you create it.

What the assistant can do

With a read-only keyWith a read-and-write key, also
search_streams: find prices and indicators by words ("btc rsi bybit")create_alert, update_alert
get_stream_value: the next live value of a streamset_alert_active, delete_alert
list_alerts, get_alert: conditions, state, last firedcreate_sink, update_sink, test_sink
list_sinks, get_sink (secrets never shown)set_sink_active, delete_sink
list_proxies, get_proxycreate_proxy, update_proxy
alert_history, sink_history, proxy_history, with a time rangeset_proxy_active, delete_proxy
message_placeholders: what a message can show

Conditions are written as a list the assistant fills in: a stream, an operator, a number or another stream, optional arithmetic ({op: "+", value: 200}) and how it joins the next one. The same compiler as the app checks them, and mistakes come back as plain sentences ("condition 1, value: Enter a number"), so the assistant can fix them itself.

Built for agents, not only chat

The MCP server is the control plane; AI agent sinks are the event plane. Together an agent can set its own triggers through MCP and get called through A2A or HTTP only when they fire, without running a polling loop.

Entities created through MCP are marked in the app with the name of the key that made them, so you always know what your agents set up.

Security

  • Keys are stored hashed; only their first characters are kept to tell them apart.
  • A read-only key cannot change anything; the write tools are not even listed for it.
  • 300 requests a minute per key; requests without a valid key get 401.
  • Secrets of sinks (bot tokens, signing secrets, headers) are never returned.

FAQ

What is an MCP server?

The Model Context Protocol is an open standard for connecting AI assistants to tools and data. An MCP server exposes tools; an assistant like Claude calls them when it needs to.

Does it work with claude.ai or ChatGPT connectors?

Those connectors sign in with OAuth, which is planned. Today the server works with clients that send an API key header: Claude Code, Cursor, VS Code, Windsurf and custom agents.

Can an agent see my Telegram token?

No. Secrets are never sent back, even to a read-and-write key.

Is it free?

Reading through MCP works on the Free plan with a read-only key. Keys that create and change alerts, sinks and proxies are part of Pro, which is free during the pilot.

Market alerts your AI assistant can set up

Free to start, Pro free during the pilot. No card, no exchange keys.

Create an alert