MCP servers aur clients
Samjhao ki Model Context Protocol AI apps ko tools aur data se kaise jodta hai, ek chhota MCP server banao, aur jaano ki kaunse security sawaal poochhne hain.
Intermediate 3 · Agent builderabout 4 hours7 sections
Seedha jawab
Model Context Protocol (MCP) AI apps ko tools aur data se jodne ka ek open standard hai: ek MCP server tools aur resources expose karta hai, aur AI apps clients ki tarah connect hote hain.
Ye kyun zaroori hai
AI assistants aur agents mein tools aur company data plug karne ka ab common tareeka MCP ban gaya hai. Clients naam leke maangte hain, jaise hamare ticketing system ke liye ek MCP server, aur FDE job posts mein bhi ye likha hota hai. Pieces pata ho toh banana tez hai.
Sign up free to read the whole lesson
The rest of this lesson: 7 parts, video lessons, the best course, practice and a quick check. It is free, takes a few seconds and needs no card.
Accounts are for learners aged 18 and over. Under 18? The quick answer and the first part of every lesson stay open: browse lessons by topic.
Lessons on Forward Deployed Engineer (FDE)0 of 90 done
- 1Python for production
- 2Project tooling: environments, tests and logs
- 3SQL for engineers
- 4REST APIs with FastAPI
- 5Files, tables and PDFs
- 6Git with a team
- 7Linux, the shell and SSH
- 8Docker and Compose
- 9Call an LLM API from code
- 10Tokens and the context window
- 11How a model picks its next word: temperature, top-p and top-k
- 12AI in Indian languages: tokens, quality gaps and what to use
- 13System prompts and prompt design
- 14Get structured JSON output
- 15Tool calling
- 16Embeddings
- 17Chunk documents for retrieval
- 18Vector search
- 19Build a RAG app
- 20Check whether retrieval works
- 21Hybrid search, reranking and filters
- 22Query rewriting, HyDE and parent-child chunks
- 23Grounding, citations and refusing to guess
- 24Long documents without vectors
- 25Test sets and evals
- 26LLM judges and their limits
- 27Choosing a model: benchmarks, leaderboards and your own tests
- 28Scoring open-ended answers: exact checks, similarity and side-by-side comparison
- 29Catching hallucinations: why models make things up and how to detect it
- 30Regression tests for prompts
- 31Quality, cost and speed trade-offs
- 32The agent loop
- 33Designing tools for agents
- 34Agent memory
- 35Planning and reflection
- 36LangGraph: agents as graphs
- 37Agent SDKs and sub-agents
- 38MCP servers and clients
- 39Context engineering
- 40Human approval steps
- 41Guardrails
- 42Permissions and sandboxing
- 43Loop limits, budgets and kill switches
- 44How agents fail
- 45Advanced harness engineering
- 46Harnesses for long-running agents
- 47Serving an AI app
- 48Kubernetes and Helm basics
- 49One cloud in depth
- 50CI/CD with GitHub Actions
- 51Environments, secrets and Terraform
- 52Deploy and monitor an LLM app
- 53Tracing LLM apps
- 54Logs and traces for agents
- 55Online evals and feedback
- 56LLM gateways
- 57Inside LLM inference: prefill, decode and the KV cache
- 58Running models on many GPUs: replicas, tensor and pipeline parallelism
- 59Grow an AI app in layers: context, guardrails, routing, caching and agents
- 60Caching and speed
- 61Cost control
- 62Batching and model tiering
- 63Latency budgets and load testing
- 64Single sign-on
- 65Permission-aware RAG
- 66Audit logs
- 67Enterprise connectors
- 68Data pipelines for AI
- 69Data quality and personal data
- 70Defend against prompt injection
- 71Security and privacy
- 72Compliance for AI systems
- 73Security questionnaires
- 74Deployment models
- 75Open-weight models with vLLM
- 76Cost of ownership
- 77Discovery
- 78The KPI that matters
- 79Writing a scope of work
- 80Proof of concept, pilot, production
- 81Change management and training
- 82Demos, risk logs and escalation
- 83Talking to executives
- 84Managing scope creep
- 85Documentation and handover
- 86A case study, end to end
- 87Design an AI system
- 88The design interview
- 89The FDE interview loop
- 90Portfolio and job search
Related
On these roadmaps: Forward Deployed Engineer (FDE), Agentic AI Engineer
AgentsIntermediate 3: Agent builderVideo lessonsModel Context ProtocolMCP serverTool callingLeast privilege