Replit Opens Its MCP Server to External AI Clients
Replit’s production-ready MCP server lets assistants including ChatGPT and Claude create, modify and deploy applications.
Replit moves beyond its own workspace
Replit has taken its Model Context Protocol server out of beta, allowing compatible AI clients to direct Replit Agent without requiring users to work inside Replit’s main interface. The company says people can now create new applications, update existing projects and manage deployments from environments such as ChatGPT, Claude and Slack.
The server uses OAuth 2.0 authentication and the streamable HTTP version of MCP. Replit’s documentation also lists support for custom clients and software-development kits that implement the protocol. Once connected, an external assistant can invoke Replit’s application-building capabilities while Replit continues to provide the underlying project environment, agent and deployment infrastructure.
This release changes the role of Replit from a self-contained AI development product into an execution service that other assistants can call. A user discussing a product idea in one conversational tool can hand the implementation to Replit Agent without copying the specification into a separate workspace. Existing applications can likewise be revised through whichever MCP-compatible interface a team already uses.
The broader update announced alongside MCP included conversational project analytics and optional nightly database snapshots. Those additions make it possible to ask an agent about traffic, funnels and experiments, while adding a separate recovery layer above the point-in-time protection already attached to production databases.
Why it matters
Coding-agent competition is increasingly about who owns execution after a user expresses an intention. By accepting instructions from rival assistants, Replit is betting that its application runtime, deployment system and project context can remain valuable even when another company controls the conversation. The approach could widen Replit’s reach, but it also makes authorization boundaries more important: an assistant able to modify and deploy an application needs narrowly scoped access, visible action records and dependable confirmation for consequential changes.