Dreamlit

What Dreamlit MCP server does

The Dreamlit MCP server gives AI clients product-level control over notification workflows in Dreamlit. Dreamlit is described as a database-driven notification workflow platform for Supabase and Postgres applications. The server can help an agent identify the correct workspace and project, find existing workflows, create or revise drafts, apply saved brand-kit context to generated email drafts, and inspect preview or builder URLs.

Workflow authoring is separate from activation. A newly created or updated draft is not live until the publishing flow has been completed. Publishing and scheduling require explicit confirmation, while unpublishing disables live triggers or schedules.

How it works

Dreamlit MCP server is available as a hosted remote MCP endpoint using Streamable HTTP:

https://mcp.dreamlit.ai/mcp

A compatible MCP client connects to this URL and begins an OAuth authorization flow. The user selects the Dreamlit workspace the client may access and approves the requested scopes. Clients without OAuth support can use a personal access token instead. OAuth access is workspace-scoped; personal access tokens are project-scoped and are created in Dreamlit project settings.

The recommended sequence starts with get_status, which returns workspace context, project setup information, schema hints, guidance, workflow state, and application URLs. Agents can then resolve project and workflow names, inspect drafts, gather analytics, write or update a draft, test it, and publish only after confirmation.

Setup and configuration

Add the production endpoint through the remote-server configuration flow of the MCP client. For Codex CLI, the documented commands are:

codex mcp add dreamlit --url https://mcp.dreamlit.ai/mcp
codex mcp login dreamlit

Codex can request the supported scopes explicitly with --scopes workflows:read,workflows:write,workflows:publish,analytics:read. Request only the permissions needed by the client: workflows:read, workflows:write, workflows:publish, and analytics:read are available.

The MCP connection does not require database credentials. Integration credentials should be configured in Dreamlit application settings, not pasted into prompts. The MCP Inspector can also connect by selecting Streamable HTTP and entering the production URL.

Tools and capabilities

The Dreamlit MCP server exposes these product-level operations:

• Discover accessible projects, workflows, and saved brand kits with pagination.
• Inspect workflow drafts and retrieve preview or builder URLs.
• Query bounded notification, recipient-engagement, and workflow-run analytics with filters and cursors.
• Create a workflow draft or update an existing draft from an outcome-oriented natural-language request.
• Send a confirmed email or Slack test without publishing the workflow.
• Validate a workflow, then publish or schedule it after explicit confirmation.
• Disable active triggers or schedules with unpublish_workflow.

Prompts should provide the triggering event or schedule, audience, message goal and tone, personalization data, timing and timezone, business rules, and unsubscribe requirements where relevant.

Limitations and notes

This is a hosted remote server; the repository contains setup documentation and registry metadata, not the production backend implementation. It exposes workflow and analytics operations rather than low-level graph editing or database APIs. Analytics results are structured, bounded, and paginated; CSV exports and bulk dumps are not available through MCP.

The server can access workflow metadata, saved brand-kit summaries, connected project context, and bounded analytics needed for requested actions. Keep raw credentials, secrets, and private tokens out of prompts. Publishing remains a distinct, confirmation-gated action after draft authoring and testing.