General Input MCP server for Claude and ChatGPT

Connect once from Settings, then MCP, and Claude, ChatGPT, Claude Code, Codex CLI, or any other MCP client can work in your General Input workspace. The connector isn't a read-only window or a short list of popular actions: it mounts the full toolset our own agent uses, so the model in your AI tool can call any operation across your connected integrations, build and edit workflows, apps, and skills, run them, and read their traces. Read the MCP announcement.

How it works

• Per-person sign-in. The server URL names your workspace, and everyone signs in with their own account through OAuth, so each person gets exactly their own access.
• One toolset, two surfaces. The connector registers the same tools the in-product agent runs, so the two can't drift.
• A persistent sandbox. Each member's connector session gets a warm Linux sandbox that keeps its files between calls and is released after five minutes idle. Read-only calls never start one.
• Credentials stay in the vault. Secrets are injected inside the sandbox, scrubbed from output, and every use is audit-logged with the reason the model gave.
• Interactive cards where the client supports MCP Apps, like a Connect button for a missing integration.
• Billing. Reasoning runs on your AI tool's own model; sandbox time and platform services bill to the workspace.