Dreamlit MCP
Dreamlit MCP lets AI clients create, inspect, test, publish, and unpublish Dreamlit notification workflows.
Dreamlit is a database-driven notification workflow platform for Supabase and Postgres apps. Use this MCP server when you want an AI client to help build notification workflows from outcome-oriented prompts, then review and publish them through Dreamlit.
Server
- Production MCP URL:
https://mcp.dreamlit.ai/mcp - Transport: Streamable HTTP
- Authentication: OAuth, with personal access token fallback for clients that do not support OAuth
This repository contains public setup docs and MCP registry metadata for Dreamlit's hosted remote MCP server. The production backend implementation is not open source.
Quick Start
Use your MCP client's remote server flow and enter:
https://mcp.dreamlit.ai/mcp
The client should open a Dreamlit authorization page. Choose the workspace you want the AI client to access and approve the requested scopes.
Codex CLI
codex mcp add dreamlit --url https://mcp.dreamlit.ai/mcp
codex mcp login dreamlit
MCP Inspector
npx @modelcontextprotocol/inspector
Then choose Streamable HTTP and use:
https://mcp.dreamlit.ai/mcp
Tools
Dreamlit exposes workflow-level tools, not low-level graph editing APIs.
| Tool | Scope | Purpose |
|---|---|---|
get_status | workflows:read | Get Dreamlit guidance, workspace context, project setup, schema hints, workflow state, and app URLs. |
list_projects | workflows:read | Find accessible Dreamlit projects in the approved workspace. |
list_workflows | workflows:read | Find workflows in a project by name, description, trigger type, and pagination. |
get_workflow_and_preview_url | workflows:read | Inspect a workflow draft and get the Dreamlit preview or builder URL before editing or publishing. |
create_or_update_workflow | workflows:write | Create a new workflow draft or update an existing draft from a natural-language prompt. |
send_workflow_test | workflows:write | Confirm and send a draft email or Slack test without publishing the workflow. |
prepare_publish | workflows:publish | Validate a workflow before publishing and return confirmation fields. |
confirm_publish | workflows:publish | Publish or schedule a workflow after explicit user confirmation. |
unpublish_workflow | workflows:publish | Disable live triggers or schedules for a workflow. |
Recommended Flow
- Call
get_statusfirst to understand the workspace, project setup, and prompting guidance. - Use
list_projectsandlist_workflowsif the user names a project or workflow without IDs. - Use
get_workflow_and_preview_urlbefore editing an existing workflow. - Use
create_or_update_workflowto create or update a draft. - Open the returned preview or builder URL and send workflow tests for important email or Slack messages.
- Publish only after explicit user confirmation, using
prepare_publishfollowed byconfirm_publish.
Drafts are not live after authoring. Publishing is always a separate explicit step.
Prompting Tips
Ask for the outcome, not a workflow graph. Good prompts include:
- The event or schedule that should start the workflow.
- The audience or recipient.
- The message goal and tone.
- The data needed for personalization.
- Timing, timezone, and business rules.
- Whether recipients should be able to unsubscribe.
Examples:
When a new organization row is inserted, send an internal Slack alert with the organization name, owner email, and plan.
Every Monday at 9 AM America/New_York, send admins a weekly usage digest summarizing active users, failed payments, and new signups.
Tomorrow at 10 AM, announce the launch to active users who opted into product updates. Include unsubscribe.
Access Model
OAuth access is scoped to the workspace selected during authorization. Personal access tokens are project scoped and can be created from Dreamlit project settings.
Available scopes:
workflows:readworkflows:writeworkflows:publish
Use the least privilege that fits the client. Read-only clients should request only workflows:read.
Data Handling
Dreamlit MCP can access workflow metadata and connected project context needed to perform the requested workflow actions. It does not require clients to send database credentials through MCP.
Do not paste secrets, raw credentials, or private tokens into prompts. Use Dreamlit's app settings for integrations and credentials.
Support
For support, contact support@dreamlit.ai.