NEW MCP

Crosslybot for AI agents

Connect Crosslybot to Claude Desktop, Cursor or any MCP client. The AI agent publishes posts to your channels in response to natural-language commands.

Hosted server at mcp.crosslybot.com — no local install

What MCP is

Model Context Protocol — an open standard from Anthropic for connecting AI agents to external tools. When you connect the Crosslybot MCP server to your AI client, the agent gets tools for working with your channels.

Instead of copying tokens into a script or hand-crafting webhook requests, you write in the chat:

"Publish to the marketing channel: announcement about v2.0 release"

Tools

crosslybot_discover

Discover structure

Returns the list of your projects and publishing targets (Telegram channels, VK groups, Max channels) — the AI learns where it can publish.

The AI calls this at the start of a conversation to understand context.

crosslybot_publish

Publish a post

Creates a post with text, media (up to 10 files), targeted publishing to specific channels. Optional deferred pause of project or channel after publish.

Targets can be specified by name fragment — fuzzy resolve finds them.

crosslybot_pause

Set a pause

Deferred pause without publishing a post: the whole project or specific channels for N minutes. A channel pause is global — applies across all your projects.

"Pause the marketing channel for an hour" — the AI resolves the target itself.

crosslybot_resume

Lift a pause

Lifts a deferred pause early — projects, channels, or everything at once. Posts accumulated during the pause publish at 1-min intervals.

scope: project / targets / all.

How to connect

1

Create a webhook IN endpoint in the bot

Open Crosslybot, go to your project → Webhook IN. Copy the slug and Bearer token. The token is shown only once at creation. Pro+ plan required.

In the Security section, enable the "Expose names in discovery" toggle — without it the AI can only target by public_id.

2

Add the config to your MCP client

Claude Desktop: file ~/.config/Claude/claude_desktop_config.json (or %APPDATA%\Claude\claude_desktop_config.json on Windows). 

{
  "mcpServers": {
    "crosslybot": {
      "url": "https://mcp.crosslybot.com/sse/<your-slug>",
      "headers": {
        "Authorization": "Bearer crossly_live_..."
      }
    }
  }
}

Cursor and Cline use the same format in their own settings files.

3

Restart the client and start chatting

The AI agent will see 4 tools: crosslybot_discover, crosslybot_publish, crosslybot_pause, crosslybot_resume. From here just write commands:

"Show me my channels"
"Publish to the marketing channel: v2.0 release tomorrow at 10:00"
"Pause the News channel for 2 hours"
"Lift all pauses"
Self-hosted

Run it yourself

If your security policy requires it, or you want to whitelist your own IP in the webhook allowlist — run the MCP server yourself. The image is published to GitHub Container Registry, multi-arch (amd64 + arm64).

Option 1 — Docker one-liner
docker run -d --name crosslybot-mcp -p 8080:8080 \
  -e CROSSLYBOT_BASE_URL=https://wh.crosslybot.com \
  ghcr.io/antiblef/crosslybot-mcp:latest
Option 2 — docker compose
git clone https://github.com/antiblef/Crosslybot-mcp.git
cd mcp-server
docker compose up -d

After starting, point your Claude Desktop config to http://localhost:8080/sse/{slug} (or your domain through an HTTPS reverse proxy). Full docs — in the README.

Security

  • ✓ Privacy by default: channel names are hidden in discovery until you explicitly enable the toggle.
  • ✓ Optional HMAC: requests can be signed additionally — add X-Crosslybot-Hmac-Secret to headers.
  • ✓ IP allowlist: if enabled, add the MCP server IP 89.223.125.61 to the allowed list.
  • ✓ Stateless server: no configuration is stored on the MCP side — slug and token live only in your client.
  • ✓ Self-hosted: optionally run the MCP server yourself via Docker, pointing to your own CROSSLYBOT_BASE_URL.

Ready to try?

Open the bot, create a webhook IN endpoint and connect Claude Desktop in a couple of minutes.

Webhook API