dlai-mcp-server

MCP server that makes DeepLearning.AI's course catalog searchable from AI coding tools like Claude Code, Codex, and Cursor.

Ask questions about 121 courses in natural language — right from your IDE.

Setup (2 minutes, zero code)

Prerequisites

• Claude Code installed (CLI, desktop app, or VS Code extension)
• That's it. No API keys, no accounts, no dependencies.

Step 1: Add the MCP server

Open your terminal and run:

# Open your Claude Code MCP config
code ~/.claude.json        # VS Code
# OR
nano ~/.claude.json        # Terminal

Add this inside the "mcpServers" section (create it if it doesn't exist):

Option A: Local via npx (recommended)

{
  "mcpServers": {
    "dlai": {
      "command": "npx",
      "args": ["-y", "dlai-mcp-server"]
    }
  }
}

Option B: Remote hosted server

{
  "mcpServers": {
    "dlai": {
      "type": "url",
      "url": "https://dlai-mcp-server-production.up.railway.app/mcp"
    }
  }
}

Note: Remote URL MCP support varies by Claude Code version. If Option B doesn't connect, use Option A. If you already have other MCP servers configured, just add the "dlai" entry alongside them.

Step 2: Restart Claude Code

Close and reopen Claude Code (or restart the CLI). The MCP server connects automatically.

Step 3: You'll see a security prompt

Claude Code will show a message like "Allow connection to dlai-mcp-server?" — this is normal. It appears for all third-party MCP servers. The server only reads public course data from deeplearning.ai.

Click "Allow" or "Trust" to proceed. This is a one-time prompt.

Step 4: Start asking questions

> "What courses does DeepLearning.AI offer about RAG?"

That's it. You're done.

What Can I Ask? (Verified Examples)

Once connected, just talk naturally. Claude automatically picks the right tool. Every example below was tested against the live server on April 2, 2026.

Find courses

"What courses about RAG?" Returns 35 courses including Retrieval Augmented Generation (RAG), Building Multimodal Search and RAG, Building Agentic RAG with LlamaIndex, Knowledge Graphs for RAG, and 31 more.

"Find beginner courses on AI agents" Returns 17 courses: Governing AI Agents (Databricks), Evaluating AI Agents (Arize AI), Agent Skills with Anthropic, Practical Multi AI Agents with crewAI, and more.

"What does OpenAI partner with DLAI on?" Returns 4 courses: Reasoning with o1, Collaborative Writing and Coding with OpenAI Canvas, Building Systems with the ChatGPT API, ChatGPT Prompt Engineering for Developers.

"What courses does Andrew Ng teach?" Returns 13 courses: Build with Andrew, Agentic AI, AI Python for Beginners, Machine Learning Specialization, and more.

"Show me courses about prompt engineering" Returns 46 courses across beginner and intermediate levels from partners like Meta, OpenAI, Comet, and more.

"What LangChain courses are there?" Returns 5 courses: Build LLM Apps with LangChain.js, Functions Tools and Agents with LangChain, LangChain Chat with Your Data, LangChain for LLM Application Development, AI Agents in LangGraph.

"What Specializations are available?" Returns 14 specializations: PyTorch for Deep Learning, Data Analytics, Generative AI for Software Development, Data Engineering, AI for Good, Machine Learning, and more.

Explore a specific course

"What does the ChatGPT Prompt Engineering course cover?" Returns: 10 lessons | 7 code examples | 1h 30m total Lessons: Introduction (6m), Guidelines (17m), Iterative (13m), Summarizing (8m), Inferring (12m), Transforming (13m), Expanding (7m), Chatbot (12m), Conclusion (2m), Quiz.

"How many lessons in the crewAI multi-agent course?" Returns: 20 lessons | 7 code examples | 2h 43m total Covers: AI Agents overview, creating research agents, customer support automation, agent tools, event planning automation, financial analysis, job application tailoring, and more.

"Tell me about Building Agentic RAG with LlamaIndex" Returns: 7 lessons | 4 code examples | 45m total Lessons: Introduction, Router Query Engine, Tool Calling, Building an Agent Reasoning Loop, Building a Multi-Document Agent, Conclusion, Quiz.

Browse topics

"What topics are available on DeepLearning.AI?" Returns 38 topics. Top 10 by course count:

• GenAI Applications: 60 courses
• Prompt Engineering: 46 courses
• Agents: 41 courses
• RAG: 34 courses
• Generative Models: 30 courses
• LLMOps: 27 courses
• Search and Retrieval: 24 courses
• Chatbots: 23 courses
• Task Automation: 20 courses
• AI Frameworks: 20 courses

Combined filters

"Find beginner Agent courses" (topic + level filter) Returns 17 courses from Anthropic, CrewAI, Databricks, Replit, LlamaIndex, Windsurf, Arize AI, OpenAI, and more.

"Fine-tuning courses" Returns 17 courses covering post-training, GRPO, LLM serving, pretraining, vision model prompting, and more.

Edge cases

"Courses on kubernetes?" → 0 results (not in catalog) "Advanced courses?" → 0 results (DLAI only has Beginner and Intermediate levels) Invalid course slug → Clear error message: "Course not found. Use search_courses to find valid slugs."

Recommendations (Claude combines results with reasoning)

These prompts work because Claude uses the search tools and then reasons about the results:

"I'm a Python developer new to AI. Where should I start?"
"What's the fastest path to learning RAG?"
"Compare the beginner RAG courses — which one should I take first?"
"I want to learn about AI agents. Give me a 4-week learning plan."

Try It Right Now (no install)

Want to verify the server works before adding it to Claude Code? Run this in your terminal:

# 1. Check the server is alive
curl https://dlai-mcp-server-production.up.railway.app/health
# Expected: {"status":"ok","courses":121}

# 2. See server info
curl https://dlai-mcp-server-production.up.railway.app/
# Returns: server name, version, course count, tool list

# 3. Test the MCP handshake
curl -X POST https://dlai-mcp-server-production.up.railway.app/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"test","version":"0.1"}},"id":0}'
# Expected: SSE response with server capabilities and 3 tools

If all 3 return data, the server is working. Add it to your config and start asking questions.

Alternative: Run Locally

If you prefer running the server on your machine instead of using the hosted version:

{
  "mcpServers": {
    "dlai": {
      "command": "npx",
      "args": ["-y", "dlai-mcp-server"]
    }
  }
}

Requires Node.js 18+. Data is cached locally at ~/.dlai-mcp/cache/courses.json and auto-refreshes every 24 hours.

Use Cases

| Who | Example Question | |-----|-----------------| | Developer learning AI | "I know Python but nothing about LLMs. Where do I start?" | | Team lead | "Find courses to upskill my team on RAG and agents" | | Student | "What's the shortest course that covers prompt engineering?" | | Career switcher | "I'm a backend dev. What's the path to AI engineering?" | | Bootcamp instructor | "Map the DLAI curriculum to my syllabus on generative AI" | | Hiring manager | "What skills does the DLAI catalog cover? Map to my job reqs" |

How It Works

You ask a question in Claude Code
       |
       v
Claude picks the right tool automatically
       |
       v
dlai-mcp-server searches DeepLearning.AI's course catalog
       |
       v
Claude formats the answer in natural language

The server queries the same Algolia search index that powers deeplearning.ai/courses. No scraping, no hacks — just the public search API.

• 121 courses across 38 topics from 70+ partners (OpenAI, Google, AWS, Meta, etc.)
• Lesson-level data fetched on-demand when you ask about a specific course
• Always up-to-date — queries the live catalog, cached for 24h

Tools Reference

For developers building on top of this MCP server:

search_courses

| Parameter | Required | Description | |-----------|----------|-------------| | query | Yes | Search term (e.g., "RAG", "agents") | | topic | No | Filter by topic (e.g., "Agents", "Computer Vision") | | level | No | "Beginner", "Intermediate", or "Advanced" | | partner | No | Filter by partner (e.g., "OpenAI", "LangChain") | | type | No | "Short Course", "Course", or "Specialization" |

get_course_details

| Parameter | Required | Description | |-----------|----------|-------------| | slug | Yes | Course slug from search results |

Returns: full metadata, lesson list with titles/durations/types, prerequisites, learning outcomes, code example count.

list_topics

No parameters. Returns all 38 topics with course counts and top 3 example courses per topic.

FAQ

Q: Is this official? A: No. This is a community-built tool that reads public data from deeplearning.ai. Not affiliated with or endorsed by DeepLearning.AI.

Q: Why does Claude Code show a security warning? A: Claude Code warns you when connecting to any third-party MCP server. This is normal and expected. Click "Allow" — the server only reads public course metadata.

Q: Is my data being collected? A: No. The server is stateless. It doesn't log queries, track users, or store any personal data. All it does is proxy requests to DeepLearning.AI's public course catalog.

Q: How current is the data? A: Live. The server queries the same Algolia index that powers deeplearning.ai/courses. New courses appear within 24 hours (cache TTL).

Q: Can I use this with Codex or Cursor? A: Yes, if your tool supports MCP over Streamable HTTP. Use the URL: https://dlai-mcp-server-production.up.railway.app/mcp

Q: The server is down / returning errors? A: Check status: curl https://dlai-mcp-server-production.up.railway.app/health. If it's down, open an issue on GitHub.

Roadmap

• Phase 1 (shipped): Course search, lesson details, topic browsing
• Phase 2: Semantic search, learning path generation, course comparisons
• Phase 3: Code example extraction, transcript search (requires DLAI eng partnership)

Development

git clone https://github.com/gauravsurtani/dlai-mcp-server
cd dlai-mcp-server
npm install
npm run build
npm test                   # 25 tests, all passing

# Test with MCP Inspector (visual):
npx @modelcontextprotocol/inspector node dist/index.js

# Test from CLI:
bash tests/mcp-test.sh search_courses '{"query":"RAG"}'

Contributing

Issues and PRs welcome at github.com/gauravsurtani/dlai-mcp-server.

License

MIT