can1357/oh-my-pi · warning · Error

Server "${name}" was disconnected during initial connection

Error message

Server "${name}" was disconnected during initial connection

What it means

During connectServers(), after a server's connection promise resolves, the manager checks whether the server was concurrently disconnected or the manager was reset (epoch changed) while connecting. If so, it detaches and disconnects the just-created connection and throws this error naming the server. This prevents registering connections for servers that were removed mid-connect (e.g. by a /mcp reload or disconnectAll).

Source

Thrown at packages/coding-agent/src/mcp/manager.ts:576

						this.#handleServerNotification(name, method, params);
					},
					onRequest: (method, params) => {
						return this.#handleServerRequest(method, params);
					},
				});
			})().then(
				async connection => {
					// Store original config (without resolved tokens) to keep
					// cache keys stable and avoid leaking rotating credentials.
					connection.config = config;
					if (sources[name]) {
						connection._source = sources[name];
					}

					if (this.#epoch !== connectionEpoch || this.#pendingConnections.get(name) !== connectionPromise) {
						this.#detachConnection(name, connection);
						void disconnectServer(connection).catch(() => {});
						throw new Error(`Server "${name}" was disconnected during initial connection`);
					}

					this.#pendingConnections.delete(name);
					this.#connections.set(name, connection);
					this.#serverConfigs.set(name, config);

					// Wire auth refresh for HTTP-like transports so 401s trigger token refresh.
					// Gate on a resolvable managed credential, not on the auth block:
					// definition-only configs (url-keyed fallback) get Bearer injection
					// too and need the same mid-session refresh hook.
					if (
						isAuthRefreshableMCPTransport(connection.transport) &&
						lookupMcpOAuthCredential(this.#authStorage, config)
					) {
						connection.transport.onAuthError = async () => {
							const refreshed = await this.#resolveAuthConfig(config, { forceRefresh: true });
							if (refreshed.type === "http" || refreshed.type === "sse") {
								return refreshed.headers ?? null;

View on GitHub (pinned to 9690622007)

Solutions

1. Retry connectServers() after the concurrent operation (reload/disconnect) completes.
2. Avoid triggering /mcp reload or disconnectAll while initial connections are in flight; wait for connection to settle first.
3. Don't issue overlapping connectServers() calls for the same server; await the first call.
4. Treat the error as a signal the server was removed from config — verify it's still in the current config before reconnecting.
5. Check logs for which concurrent operation raced the connect.

Example fix

// before
void manager.connectServers(); // fire-and-forget; user reload races it
// after
await manager.connectServers(); // await before allowing reload UI actions
enableReloadButton();

// Only connect when no conflicting lifecycle operation is in flight
if (manager.isReloading() || manager.isDisconnecting(name)) {
  await manager.waitForSettled(); // or defer until reload completes
}
await manager.connectServers();

null

try {
  await manager.connectServers();
} catch (e) {
  if (e instanceof Error && /was disconnected during initial connection/.test(e.message)) {
    // server removed/reloaded mid-connect; retry once after settling
    await Bun.sleep(250);
    return manager.connectServers();
  }
  throw e;
}

When it happens

Trigger: Calling connectServers() while, for the same server name: disconnectServer/disconnectAll runs, the manager epoch is bumped (reload), a duplicate connect for the same name supersedes the pending promise, or #detachConnection is invoked by another path during the handshake.

Common situations: User runs /mcp reload while servers are still connecting at startup; a slow-connecting server (slow stdio spawn or remote OAuth flow) finishes after the config was reloaded; two overlapping connectServers() calls racing for the same name; manager reset during shutdown while connections still initializing.

Related errors

• MCP server not connected: ${name}
• Server "${name}" was disconnected during reconnection
• Cannot set cwd on a disposed JS runtime
• Transport not connected
• ${name}: ${message} (joined by "; ")

AI-assisted analysis of can1357/oh-my-pi@9690622007 (2026-08-31). Data as JSON: /api/errors/ef09c4dcef285013. Report an issue: GitHub.