Dreamlit MCP Server

Content

Dreamlit MCP Server connects MCP-capable AI clients to Dreamlit's hosted notification workflow platform for Supabase and Postgres applications. It is designed for outcome-oriented workflow authoring: create drafts from prompts, inspect existing workflows, send tests, apply saved brand kits, publish or unpublish workflows, and answer bounded analytics questions about notification performance.

The production server endpoint is:

https://mcp.dreamlit.ai/mcp

Dreamlit documents the transport as Streamable HTTP. Authentication is OAuth, with a personal access token fallback for clients that do not support OAuth.

Source Review

• https://github.com/dreamlit-ai/dreamlit-mcp
• https://modelcontextprotocol.io/registry/about
• https://code.claude.com/docs/en/mcp

These sources were reviewed on 2026-06-05. Prefer the live Dreamlit repository and current MCP client documentation over model memory for endpoint behavior, scopes, token handling, and tool coverage.

Features

• Hosted remote MCP endpoint at https://mcp.dreamlit.ai/mcp.
• Streamable HTTP transport for compatible MCP clients.
• OAuth workspace authorization, with project-scoped personal access token fallback for clients that do not support OAuth.
• Workflow discovery through project and workflow listing tools.
• Draft authoring and updates from natural-language workflow prompts.
• Brand kit lookup before applying saved styling to generated email drafts.
• Preview and builder URLs for human review before publishing.
• Workflow test sending without making a workflow live.
• Two-step publishing flow with prepare_publish followed by confirm_publish.
• Bounded, filtered, cursor-paginated notification analytics access.

Tools

• get_status: read Dreamlit guidance, workspace context, project setup, schema hints, workflow state, and app URLs.
• list_projects: find accessible projects in the approved workspace.
• list_workflows: find workflows by name, description, trigger type, and pagination.
• list_brand_kits: inspect saved brand kit IDs and style summaries.
• get_workflow_and_preview_url: inspect a draft and get a preview or builder URL before editing or publishing.
• get_analytics: query bounded recipient engagement and workflow run data.
• create_or_update_workflow: create a new draft or update an existing draft.
• send_workflow_test: confirm and send a draft email or Slack test.
• prepare_publish: validate a workflow and return confirmation fields.
• confirm_publish: publish or schedule a workflow after explicit user confirmation.
• unpublish_workflow: disable live triggers or schedules.

Installation

Codex CLI

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

To request every documented v1 scope explicitly:

codex mcp login dreamlit --scopes workflows:read,workflows:write,workflows:publish,analytics:read

For a narrower read-only setup, request only the scopes needed for the client, such as workflows:read or analytics:read.

Generic Remote MCP Client

Use the client's remote server flow and enter:

https://mcp.dreamlit.ai/mcp

The client should open a Dreamlit authorization page. Choose the workspace the AI client may access, then approve only the requested scopes that match the workflow.

Access Model

Dreamlit OAuth access is scoped to the workspace selected during authorization. Personal access tokens are project scoped and can be created from Dreamlit project settings.

Documented scopes:

• workflows:read
• workflows:write
• workflows:publish
• analytics:read

Use the smallest scope set that fits the job. A review-only assistant generally needs workflows:read; analytics-only tasks should request analytics:read; authoring and publishing should be separated when possible.

Recommended Workflow

1. Call get_status first to understand workspace context and project setup.
2. Use list_projects and list_workflows when the user gives names instead of IDs.
3. Use list_brand_kits before applying a saved style to generated email drafts.
4. Use get_workflow_and_preview_url before editing an existing workflow.
5. Use get_analytics only for bounded questions, with filters and cursor pagination.
6. Create or update a draft with create_or_update_workflow.
7. Open the returned preview or builder URL and send tests for important email or Slack messages.
8. Publish only after explicit confirmation, using prepare_publish followed by confirm_publish.

Data Handling

Dreamlit MCP can access workflow metadata, saved brand kit summaries, connected project context, preview URLs, and bounded analytics needed for the requested action. The public docs state that clients do not need to send database credentials through MCP.

Avoid putting secrets, raw credentials, private tokens, or integration credentials in prompts. Use Dreamlit's app settings for integrations and credential management.

Troubleshooting

The client does not open Dreamlit authorization

Confirm the client supports remote Streamable HTTP MCP servers and OAuth. If it does not, use Dreamlit's documented personal access token fallback.

The assistant cannot see a workflow

Check that the authorized workspace and project are the ones the user expects. Then verify that the granted token includes workflows:read.

A draft exists but is not live

That is expected. Dreamlit's documented flow keeps authoring and publishing separate. Run tests, open the preview or builder URL, then publish only after a human confirms the workflow.

Analytics requests return incomplete results

Use narrower filters and follow returned cursors for larger result sets. The Dreamlit docs describe analytics access as structured and paginated rather than bulk CSV export.