Audience
- Developers building AI applications beyond a single model call.
- Automation engineers introducing stateful or tool-connected workflows.
- Technical architects evaluating orchestration and integration choices.
AI & technology
LangChain, LangGraph & MCP
Explore frameworks and protocols for connected, stateful AI applications.
Learning with a practical purpose.
Develop a clear understanding of three related but different building blocks for AI applications. LangChain provides abstractions for working with models and tools; LangGraph supports explicitly structured, stateful workflows; Model Context Protocol provides a standard way for AI applications to connect with external capabilities. The module explores how these fit together without assuming they must all be used in every solution.
You will build a small research-and-review assistant whose progress can be inspected and resumed. Practical sessions connect tool contracts, workflow state, persistence and human decisions, then introduce an MCP connection with deliberately limited capabilities. The outcome is an understandable application architecture and evidence that its important paths behave as intended.
Live onlineTeaching through Kent Business College, UK
2 hoursOne live session each teaching week
14 sessionsA semester for each module
28 live hoursPer module, plus independent study
The outline below is a 14-session online module, with one two-hour live session each teaching week. Reading weeks, breaks and additional activities follow the agreed cohort calendar. The cohort syllabus, full route, dates, fees, independent study and assessment requirements are confirmed before enrolment.
Who this module is designed for.
Starting knowledge and preparation
- Working Python or JavaScript skills, including functions and asynchronous API calls.
- Familiarity with JSON, environment variables and version control.
- Prior introductory work with language models and tools is recommended.
What you will work towards.
- Explain the distinct responsibilities of LangChain, LangGraph and MCP.
- Build a model-and-tool workflow with explicit data structures and error handling.
- Represent application progress as graph state with conditional transitions.
- Use persistence and human interruption to support reviewable, resumable execution.
- Connect a limited MCP capability and test the resulting end-to-end application.
Explore the 14-session learning plan.
Explore the focus and practical output of each teaching session. Open individual sessions below, or expand the complete plan.
Session 01 Choosing an orchestration approach
Map a research-and-review task and decide which parts need fixed logic, model judgement or external tools.
Practical output: An architecture decision and workflow sketch.
Session 02 Models, messages and structured results
Connect a model through the chosen language environment and validate the structure of its returned output.
Practical output: A reproducible model-call example.
Session 03 Tool definitions and contracts
Define a small tool with explicit inputs, outputs and errors that can be tested independently.
Practical output: A tested tool interface.
Session 04 Composing an initial application
Combine model, tool and validation steps while keeping important business rules outside ambiguous natural-language instructions.
Practical output: A simple application pipeline.
Session 05 Graph state and nodes
Represent the workflow’s progress explicitly and define the responsibility and state changes of each node.
Practical output: A typed or validated state model.
Session 06 Edges, routing and termination
Add conditional transitions, completion criteria and bounded loops to handle alternative paths through the task.
Practical output: A graph with tested routing cases.
Session 07 Persistence and checkpointing
Save workflow state and inspect what information is needed to continue after a pause or interruption.
Practical output: A checkpointed workflow experiment.
Session 08 Human review and resumption
Introduce a review point where a person can approve, amend or reject the proposed next action.
Practical output: A resumable review workflow.
Session 09 Errors, retries and external effects
Separate repeatable computation from consequential actions and test recovery without duplicating an external change.
Practical output: A failure-recovery test record.
Session 10 MCP architecture and capabilities
Distinguish host, client and server roles and explore how tools, resources and prompts are exposed.
Practical output: An MCP capability map.
Session 11 Connecting a limited MCP service
Integrate a controlled server capability with clear permissions and inspect the request and response boundaries.
Practical output: A working restricted MCP connection.
Session 12 Tracing and behavioural evaluation
Review execution paths and test ordinary, ambiguous and failed requests against declared acceptance criteria.
Practical output: An evaluation suite and trace review.
Session 13 Research-and-review integration
Combine the graph, persistence, tools and MCP connection into a coherent application with documented configuration.
Practical output: An integrated research assistant.
Session 14 Architecture demonstration and defence
Demonstrate interrupted execution and recovery, then justify the components included and those deliberately left out.
Practical output: A final architecture review and handover.
Apply the learning
A resumable research-and-review assistant
Create an assistant that gathers approved information, prepares a structured briefing and pauses for a reviewer before completion. Use explicit graph state and one limited MCP capability. Demonstrate a successful run, a rejected proposal and recovery from a failed step, explaining how repeated execution avoids unintended duplicate actions.
Your project evidence can include
- A repository with configuration and setup instructions.
- A workflow graph, state model and tool contracts.
- Execution traces, evaluation cases and a component-selection rationale.
Use anonymised, public or suitable synthetic material where workplace information is confidential. The issued assessment brief confirms the required evidence.
Assessment, feedback and certification.
- Practical checkpoints assess state modelling, tool contracts and graph transitions.
- The project is tested through review, interruption and recovery scenarios.
- A technical discussion examines the architecture and the boundaries of the MCP integration.
Teaching and the relevant course certificate are provided through Kent Business College in the UK. Completion requirements are set out in the module information; attendance alone should not be assumed to meet every assessment requirement.
Questions worth asking.
Are these three names interchangeable?
No. They address application abstractions, stateful orchestration and a connection protocol respectively. Understanding their different roles is a central learning objective.
Do I need LangChain to use LangGraph?
No. LangGraph can be used independently. The module examines when combining components is useful and when a smaller design is adequate.
Is MCP an authentication policy?
MCP defines connection and interaction mechanisms. You still need an appropriate authorisation design and restricted access for the capabilities your application exposes.