Connect your AI with MCP
The @upon-ai/mcp-server package connects MCP-compatible AI clients to the
UponAI public API. Requires Node.js 22.14 or later. Licensed under MIT by UponAI.
Configure your MCP client
Add this entry to your client's MCP server configuration. Replace the token and workspace ID with your own values. Use the client's secret store when available; do not commit credentials to source control.
{
"mcpServers": {
"uponai": {
"command": "npx",
"args": ["-y", "@upon-ai/mcp-server@0.1.1"],
"env": {
"UPONAI_API_KEY": "YOUR_CUSTOMER_API_TOKEN",
"UPONAI_WORKSPACE_IDS": "123"
}
}
}
}
Restart or reconnect the MCP client, then ask it to list agents in your configured workspace. The server runs locally over stdio; no hosted MCP URL is required.
Find your workspace ID using GET /api/v1/workspaces, the SDK's
client.listWorkspaces(), or the MCP workspace-discovery tool. Without
UPONAI_WORKSPACE_IDS, only discovery is enabled. For multiple workspaces, use
an explicit comma-separated list such as 123,456.
Available tools
Workspace discovery plus 18 curated read tools cover agents, LLMs, conversation flows, chats, calls, knowledge bases, voices, and concurrency. Responses contain limited summaries and exclude secrets, full transcripts, recordings, and raw agent configuration. API authorization still applies to every request.
Write tools are disabled by default. Advanced operators can explicitly enable
createPhoneCall, endLiveCall, or rerunCallAnalysis through
UPONAI_MCP_WRITE_OPERATIONS. Each write also requires confirm: true and an
allowed workspace. Configure your AI client's approval controls before enabling
writes, which may incur charges or affect live calls.
Optional self-hosted HTTP transport
For an operator-managed deployment, the package also supports Streamable HTTP with separate MCP authentication. This is a single-principal deployment; npm installation does not create a hosted service or delegated customer login. See the package README for listener, proxy, and credential configuration.