Open-source AI teammates in Slack, Teams, Discord, Telegram and Mattermost. Self-host it, with any agent harness and any model.
Python agent-orchestrationaiai-agentschatopsdiscordlangchainmcpmicrosoft-teamsmulti-agentopen-sourceopenaiself-hostedslack
import { randomBytes } from 'node:crypto';
import { serveMcpOverHttp } from '@sandboxaq/switch-agent-runtime/hosted';
import { z } from 'zod';
import type { HttpMcpServerSpec } from '../adapter';
import type { ParentChannel } from './session-channel';
const toolsSchema = z.array(
z.object({
name: z.string().min(1),
description: z.string(),
inputSchema: z.record(z.string(), z.unknown()),
})
);
const toolResultSchema = z.object({
isError: z.boolean().optional(),
content: z.array(z.object({ type: z.literal('text'), text: z.string() })),
structuredContent: z.record(z.string(), z.unknown()).optional(),
});
/**
* The Switch tools, served to this session's CLI on loopback.
*
* Bound before the CLI starts, on a port the system picks, and answering only
* the bearer token minted here for this run; a restart gets a new port and a
* new token. Nothing is decided here: the tool list and every call go up the
* IPC pipe to the parent, which holds the agent's connection and is the only
* thing that talks to Switch. `spec` is what the CLI is given.
*/
export async function startSessionMcp(
parent: ParentChannel
): Promise<{ spec: HttpMcpServerSpec; close: () => Promise<void> }> {
const token = randomBytes(32).toString('hex');
const server = await serveMcpOverHttp(token, {
listTools: async () => toolsSchema.parse(await parent.ask({ type: 'tools' })),
callTool: async (name, args) =>
toolResultSchema.parse(await parent.ask({ type: 'tool', name, arguments: args })),
});
return {
spec: { transport: 'http', url: server.url, headers: { Authorization: `Bearer ${token}` } },
close: server.close,
};
}