Quickstart - AdStellar

Claude Desktop / ChatGPT (OAuth — recommended)

Add AdStellar to your claude_desktop_config.json with no key at all:

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

On first use the client registers itself, opens the AdStellar consent screen in your browser, and you pick the team + scopes to grant. Tokens refresh automatically.

API key (for your own agents)

Create a key in Settings → Developers (or via POST /api/settings/api-keys), then:

{
  "mcpServers": {
    "adstellar": {
      "url": "https://app.adstellar.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer ast_live_YOUR_KEY"
      }
    }
  }
}

Restart the client. You should see AdStellar’s tools — try:

“What’s my current credit balance and which ad accounts are connected?”

“Generate 5 headline variants for my default brand kit, bold tone, goal: drive trial signups.”

“Generate a 9:16 image ad for my product, 2 variations” — then “check the status of batch <id>”

“How did my campaigns perform over the last 7 days, grouped by campaign?”

REST API

Every MCP tool is also a REST endpoint — POST /api/v1/tools/{toolName} with the same bearer credential:

curl -X POST https://app.adstellar.ai/api/v1/tools/get_credit_balance \
  -H "Authorization: Bearer ast_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'

The full spec lives at the API Reference tab (or GET /api/v1/openapi.json).

Raw JSON-RPC (for debugging)

# Discover tools
curl -X POST https://app.adstellar.ai/api/mcp \
  -H "Authorization: Bearer ast_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# Call a tool
curl -X POST https://app.adstellar.ai/api/mcp \
  -H "Authorization: Bearer ast_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": { "name": "get_credit_balance", "arguments": {} }
  }'

Supported methods: initialize, ping, tools/list, tools/call (single or batched). The server is stateless — every POST is self-contained.

Building your own agent

Create a key with only the scopes your agent needs.
On startup, call tools/list to discover the tools visible to that key.
Call get_team_info to confirm which team you’re acting on.
Use list_models before any generation call so you can pass an explicit model id (or omit it for the default router).
Generation is async: call generate_image/generate_video, then poll get_generation_status until complete: true.

​Claude Desktop / ChatGPT (OAuth — recommended)

​API key (for your own agents)

​REST API

​Raw JSON-RPC (for debugging)

​Building your own agent

Claude Desktop / ChatGPT (OAuth — recommended)

API key (for your own agents)

REST API

Raw JSON-RPC (for debugging)

Building your own agent