AI Agents and MCP

Pushmark supports both direct API calls and AI agent workflows.

That gives you two good integration paths:

• use HTTPS directly from scripts, CI/CD jobs, and backend services
• use MCP from AI agents such as Codex, Cursor, Claude Code, VS Code, and other MCP-compatible clients

If you already know the exact request you want to make, HTTP is the simplest option. If you want an agent to inspect channels, check usage, and send notifications through structured tool calls, MCP is the better fit.

CLI vs MCP

Use direct HTTP or CLI when:

• you are wiring Pushmark into a shell script
• you are sending notifications from GitHub Actions or another CI system
• you want the smallest possible integration surface

Use MCP when:

• an AI agent should choose which Pushmark action to take
• you want the model to work with structured tools instead of handwritten API glue
• you want one Pushmark integration that can be reused across multiple agent clients

CLI examples

cURL

curl -X POST https://api.pushmark.app/YOUR_CHANNEL_HASH \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Production deploy completed successfully.",
    "type": "success"
  }'

Pushmark CLI

pushmark -t success YOUR_CHANNEL_HASH "Production deploy completed successfully."

Install guide: pushmark/pushmark-cli

These examples are ideal for terminal usage, deploy hooks, cron jobs, and lightweight automations.

MCP overview

Pushmark exposes a remote MCP server at:

https://api.pushmark.app/mcp

The MCP server lets AI agents work with Pushmark as a tool-based integration instead of a prompt-only integration.

Public workflows can be used with a channel_hash. Owner workflows can use OAuth and bearer tokens for account-level actions.

Supported MCP tools

Tool	Purpose	Auth
pushmark.server_status	Check auth mode and supported scopes	No
pushmark.channel_overview	Inspect a channel, subscriber counts, and recent notifications	No with channel_hash; Yes for owned lookup
pushmark.send_notification	Validate or send a notification	No with channel_hash; Yes for owned lookup
pushmark.list_channels	List your channels	Yes
pushmark.usage_overview	Inspect limits and account usage	Yes
pushmark.create_channel	Create a channel	Yes
pushmark.update_channel	Update a channel	Yes

AI agent examples

Once Pushmark is connected over MCP, you can give your agent instructions like:

• "Send a success notification to channel YOUR_CHANNEL_HASH saying the deploy is live."
• "Show me the recent activity for channel prod-alerts."
• "List my Pushmark channels and tell me which one was used most recently."
• "Check whether my account is close to its monthly notification limit."
• "Create a new public channel called deployments-eu."

Client setup

Install Pushmark in your client

Deeplink where supported, command fallback otherwise.

Cursor

Install

VS Code

Install

Codex

Claude Code

Codex

codex mcp add pushmark --url https://api.pushmark.app/mcp
codex mcp list

Or add it directly to ~/.codex/config.toml:

[mcp_servers.pushmark]
url = "https://api.pushmark.app/mcp"

Suggested agent instruction:

Use the `pushmark` MCP server for all Pushmark operations. Complete the browser login flow if prompted.

Cursor

Create ~/.cursor/mcp.json:

{
  "mcpServers": {
    "pushmark": {
      "url": "https://api.pushmark.app/mcp"
    }
  }
}

Useful commands:

cursor-agent mcp list
cursor-agent mcp login pushmark
cursor-agent mcp list-tools pushmark

Claude Code

claude mcp add --transport http --scope user pushmark https://api.pushmark.app/mcp
claude mcp list

Claude Code will usually trigger the browser login flow automatically when an authenticated Pushmark tool is used for the first time.

Auth model

• Public channel workflows can work with channel_hash
• Account-level workflows require OAuth
• Authenticated MCP clients can list channels, inspect usage, and manage channels

If your AI agent only needs to send notifications into a known public channel, the unauthenticated path is often enough.