MCP Servers Explained: Hosting MCP Servers on Posit Connect | Workflow Demo Series
Model Context Protocol (MCP) servers provide AI assistants such as Posit Assistant and Claude with a secure, repeatable way to access your organization's tools and data. In this workflow demo, Matt Leary (Product Manager, Posit Connect) and Samiul Haque (Senior Solutions Advisor, Posit) explain how MCP works, when a custom MCP server makes sense, and how to build, deploy, and use MCP servers on Posit Connect. Topics covered in this video: - How MCP clients, servers, and tools fit together - When to build an MCP server, and when a skill is the better fit - Writing an MCP server in Python with FastMCP - Deploying from Positron in one click with Posit Publisher - Testing tools in the browser with Connect's built-in MCP landing page - Signing in with OAuth 2.1 (no API keys to manage) and per-user Snowflake access through viewer integrations - Calling the same server from Posit Assistant and Claude Desktop - Life sciences examples: adverse event classification, clinical study summaries, and a CDISC SDTM validator - Monitoring who uses your MCP servers Relevant Timestamps: 00:00 Introduction and agenda 01:10 What is MCP? Protocols, servers, and clients 02:44 Local vs. remote servers, and MCP vs. skills 03:53 What's behind an MCP server 05:13 What to weigh before you build one 09:03 Hosting remote MCP servers on Posit Connect 11:13 Connecting to an MCP server on Connect with OAuth 12:45 Building an MCP server with FastMCP 14:55 Example: Snowflake MCP server for clinical data 16:10 One-click deploy with Posit Publisher 16:34 The MCP tab and interactive landing page on Connect 19:18 Per-user Snowflake access with viewer integrations 22:22 Custom landing page: CDISC SDTM validator 23:10 Using MCP tools in Posit Assistant 25:54 Classifying adverse events from your loaded data 28:03 Connecting from Claude Desktop 31:04 Monitoring MCP server usage 32:22 Q&A: Deploying R-based MCP servers with mcptools 33:36 Q&A: Memory footprint and minimum processes 35:14 Q&A: Amazon Bedrock and IAM roles 36:09 Q&A: OAuth with Posit Workbench 37:07 Q&A: Custom vs. third-party MCP servers 40:10 Q&A: Microsoft Copilot 40:53 Q&A: Posit Assistant and PHI 42:45 Q&A: Supported languages (Python, R, Node.js) Resources and additional links: MCP servers on Posit Connect (docs): https://docs.posit.co/connect/user/mcp-servers/ Why Connect is a great place to host your MCP servers (blog, part 1): https://posit.co/blog/hosting-mcp-servers-on-connect MCP servers on Connect: Managing credentials and access (blog, part 2): https://posit.co/blog/mcp-servers-on-connect-managing-credentials-and-access Publishing from Positron or VS Code with Posit Publisher: https://docs.posit.co/connect/user/publishing-positron-vscode/ Viewer integrations on Connect (Snowflake, Databricks, and more): https://docs.posit.co/connect/user/oauth-integrations/ Posit SDK for Python, Snowflake integration: https://posit-dev.github.io/posit-sdk-py/reference/connect.external.snowflake.html Posit Assistant: https://positron.posit.co/assistant.html Connecting MCP servers to Posit Assistant: https://assistant.posit.co/docs/reference/mcp-servers/ mcptools (MCP for R): https://posit-dev.github.io/mcptools/ FastMCP: https://gofastmcp.com/ Model Context Protocol: https://modelcontextprotocol.io/ Download Positron: https://positron.posit.co/download.html Learn more about Posit Connect: https://posit.co/products/enterprise/connect Talk to our team: https://posit.co/schedule-a-call This video is part of Posit's Monthly Workflow Demo series. #mcp #positconnect #datascience