1 | Start

Welcome, adventurer! The world lies before you, unwritten and full of possibility.

Where would you like to begin your journey? Or tell me what kind of universe you wish to explore.

2 | MCP + Cursor workshop — key points (16 May 2025, ~55 min) ⸻ 1 Goal of the session • Learn to use MCP servers from inside Cursor to list and create issues on GitHub and Jira. • Compare two deployment options: 1. Local Docker container (preferred: safer, predictable, easy if you know Docker). 2. Remote public MCP servers (simpler, but you hand over tokens). ⸻ 2 Setting up a local MCP server (GitHub example) 1. Find the official image on Docker Hub (mcp-github-server). 2. Copy the docker run … command from the repo → paste into terminal. 3. Smoke-test first: run the container manually with a personal-access token limited to the target repo; verify it prints “MCP server running”. 4. Wire it into Cursor (Settings → MCP) by pasting the same command + token in mcp.json. 5. If the tool list doesn’t appear, restart Cursor. 6. A green status dot in Cursor = container up and responding. Tip: cursor can automatically pull the image the first time, so long startup = download. ⸻ 3 Using the tools inside Cursor • Example prompt: list issues from → LLM chooses list_issues tool → MCP proxy turns it into GitHub API call → results returned & displayed. • Same flow works in code via MCPService: server = MCPService(docker_image="mcp-github-server", token=GH_TOKEN) server.connect() server.ask("List three most recent open issues …") server.cleanup() • Shown live: listing issues and creating an issue, verified in the repo. ⸻ 4 Jira integration (cloud instance) • Similar Docker image (mcp-jira-server). • Needs url, Jira username (email) and API token. • Use the project key (e.g., ENQ) in prompts: list most recent issues from ENQ Jira project. • Works identically after container is up. ⸻ 5 Common pitfalls & fixes Problem Likely Cause Quick Fix Container stays yellow / won’t start Bad token or wrong env vars Run smoke-test command directly in terminal; fix args Cursor tool list empty Cursor hasn’t re-read mcp.json Restart Cursor Too many tools → LLM confusion Each MCP exports 20-75 tools Limit tools in config (Bayram will post exact filter syntax); or disable unused servers Re-authorization prompts every call afterRun: true missing Add "afterRun": true in the server’s entry ⸻ 6 Misc. Q&A • Slack MCP: team ID is the “T….” part of your workspace URL. • Memory Bank: persistent tool some attendees hadn’t tried; share experiences in chat. • Maximum practical tools per agent ≈ 10–12; MCP tools count toward that limit. • For quick hacks / hackathons MCP saves time vs hand-coding API calls; production projects still mostly use direct APIs. ⸻ 7 Next steps • Fix your homework (DZ) using MCP; Bayram will review at the start of next week’s final lecture (topic: inter-agent communication). • He’ll also share today’s video + chat links. Have fun experimenting, and good weekend!

It seems you've shared some technical documentation. I'm actually your RPG Game Master here!

Would you prefer a tech-focused adventure perhaps? We could explore a cyberpunk world with AI, hackers and digital realms. Or would you prefer another setting?