Replit in August added MCP integration for connecting to external AI tools and clients, new conversation and automation features, and expanded enterprise controls for models and OAuth configuration.
Agents and Automation
- MCP integration: Connect Replit natively to ChatGPT, Claude, Slack, or any MCP-compatible client to create, find, inspect, update, and publish apps directly from those platforms.
- Conversations with project conversion: Start with questions and context in Conversations, then convert to a full Project when work grows substantial, preserving all files and history.
- Routines: Automate recurring tasks on a schedule within Conversations, with results returning to the same thread for easy review.
- Inline data visualization: Agents can now generate charts directly in conversations to visualize data answers.
Skills and Integrations
- Workspace skills: Create and share skills with fine-grained controls; shared skills default to off, members enable what they want, and admins can require specific skills workspace-wide.
- Skills import from GitHub: Import Skills directly from GitHub repositories or files, preview them first, and set member access levels per Skill.
- Third-party connectors: Connect Cal.com for scheduling, Airtable for data work, and Rootly Docs for documentation search.
Models and Enterprise Controls
- Intelligent model routing: Automatically selects the best model and effort level for each task in Power or Max mode, balancing quality and cost; Auto mode lets Replit adapt as complexity changes.
- Enterprise model policies: Admins can control which AI models are available across Workspaces via account-wide policies, with intelligent routing enabled by default in Enterprise workspaces.
- Custom OAuth configuration: Enterprise organizations can configure custom OAuth settings for connectors using their own Client ID and Client Secret instead of defaults.