MCP training that connects your company's AI to the systems it runs on
The Model Context Protocol (MCP) lets AI apps like Claude, Cursor, Codex and ChatGPT read company data and take actions in your tools. Algoritma runs a hands-on course that teaches your leaders, business teams and engineers to use it, secure it and build on it, using two production MCP servers: Sectors for market data and Viably for day-to-day work.
The course has six modules and three tracks. We run it on site or online for teams in Indonesia and Singapore.
Which IDX banks grew CASA fastest last quarter? Then add a task for me to brief Rina Wijaya on it by Friday.
One standard plug between AI apps and your systems
An AI app on its own only knows what it learned in training. An MCP server gives it a list of things it can look up and do in one of your systems. The app is the client, your system sits behind the server, and the protocol is the same whichever app or system you use.
Tools
The model calls them
Functions with a name, a description and an input schema. The model picks one, fills in the arguments and reads the result.
create_taskViably: creates a task and posts its card to Slack
Resources
A person attaches them
Records the user pulls into the conversation before the model starts, so it works from the right context.
@viably:Viably: attaches a contact with their deals and timeline
Prompts
A person runs them
Saved workflows that chain several tools in a fixed order. In Claude Code they show up as slash commands.
/mcp__viably__triage_inboxViably: works through the inbox and proposes an action for each item
Four AI apps and four systems: count the connectors
Before MCP, every AI app needed its own integration with every system. Toggle the view to see how the work changes once each system has one MCP server.
16
Custom integrations to build and maintain. Each app has its own plugin format and auth flow, so every new AI tool adds four more connectors and every new system adds four more again.
What your company gets from MCP
MCP also gives security and operations teams the controls they need before they let agents near real data.
Build an integration once
Answers come from your data
Agents act as the person using them
People approve every change
Every call leaves a record
Switch AI vendors freely
What happens in one MCP session
Participants in every track learn the six steps between adding a server and reading a result. Engineers go on to implement each one. Click a step to read the messages.
claude mcp add --transport http viably https://api.viably.app/mcpWe teach on two MCP servers that run in production
Most companies need agents that read trusted data and agents that make changes. We teach each pattern on a server that runs in production today, so your team practises on real tools.
Sectors MCP
A hosted financial data server for Indonesian and Singaporean markets. Agents use it to look up fundamentals, filings, ownership, ESG ratings, news and commodity prices before they answer.
- More than 40 tools, with nothing to install
- Remote server over Streamable HTTP, with setup guides for Claude and ChatGPT
- Open source agent skills for common research tasks
- A usage log of every call per API key, exportable to CSV
Read the Sectors MCP guide or see our grounded agents program.
Viably MCP
A work operating system with its own CRM, task board and event ledger. Agents read the inbox and pipeline, then claim contacts, move deals and file tasks once a person allows each change.
- OAuth sign-in per person, with scopes chosen on the consent screen
- Read access is on by default. Each type of change stays off until an admin turns it on
- Saved prompts and attachable records, alongside its tools
- An audit trail that records each change and the app that made it
- No tools that send messages to customers
See the Viably MCP server, the Claude setup or our workforce automation program.
Six modules, from first connection to company-wide rollout
Each module mixes short teaching with a lab on real servers, and ends with something your team keeps. The dots show which server each module uses.
MCP foundations
What the Model Context Protocol is, the parts that make it up, and how a model decides which tool to call.
What we cover
- Hosts, clients and servers, and what each one is responsible for
- The three things a server offers: tools, resources and prompts
- Local servers over stdio and remote servers over Streamable HTTP
- How tool names, descriptions and schemas shape what the model does
Hands-on lab
Connect Claude Code or Claude.ai to the Sectors and Viably MCP servers, list their tools, and trace one request from prompt to tool call to answer.
Grounding answers in live data
Why models invent figures, and how an MCP server gives them current data to answer from instead.
What we cover
- Training data cutoffs and why a confident answer can still be wrong
- Retrieval through tool calls, and asking the model to cite each figure
- Company fundamentals, filings, ownership, ESG and news through Sectors
- Sectors' open source agent skills for repeatable research tasks
Hands-on lab
Ask the same market question with and without the Sectors server connected, then trace every number in the grounded answer back to the call that returned it.
Agents that do work
Moving from agents that answer questions to agents that change records, with a person approving each change.
What we cover
- Read tools and change tools, and how clients treat them differently
- Choosing scopes on the consent screen when an agent signs in
- Saved prompts like triage_inbox, and attaching a record with @viably:
- Where agent work shows up: the dashboard, Slack and the audit trail
Hands-on lab
Connect Viably with read scopes only, then grant tasks:write and run the inbox triage prompt, allowing or denying each change the agent proposes.
Security and governance
The controls that let a company say yes to MCP: identity, least privilege, approval and logging.
What we cover
- OAuth sign-in per person compared with shared API keys
- Least privilege scopes, and read only access for agents that run unattended
- Prompt injection through emails and form fields, and servers that label outside text as data
- Audit trails, usage logs, rate limits and the questions to ask any MCP vendor
Hands-on lab
Plant an instruction inside a test form submission and watch what an agent does with read only scopes and with write scopes. Then review a Sectors API usage log and a Viably audit trail.
Build your own MCP server
Wrapping one of your internal systems in an MCP server so every AI app in the company can use it.
What we cover
- The official MCP SDKs for TypeScript and Python
- Tool design: names, descriptions, input schemas and structured output
- Read and write annotations, and errors a model can recover from
- OAuth with your identity provider, and deploying a remote server
Hands-on lab
Build a server over one of your internal systems, or our sample dataset, then run a task where the agent uses your server, Sectors and Viably together.
Rollout and measurement
Taking MCP from a pilot group to the rest of the company, with numbers to show it worked.
What we cover
- Picking the first workflows by volume, risk and how easy they are to check
- Deciding which roles get which servers and scopes
- Onboarding staff who have never used an AI agent
- Measuring time saved and error rates, and keeping prompts and tools current
Hands-on lab
Choose three workflows, write a success measure for each, and draft a 30, 60 and 90 day rollout plan with owners.
A track for each group in your company
Leaders, business teams and engineers need different depth. Pick one track or combine them, and we adjust the content to your industry and systems.
Leadership briefing
Half dayExecutives and heads of function
What MCP changes for the business, where the risks sit, and how to pick and fund the first workflows. Includes a live demo with Sectors and Viably.
Modules 1, 4 and 6, condensed
Practitioner track
2 daysAnalysts, finance, sales, marketing and operations
Hands-on work in Claude with real MCP servers. Participants ground research in Sectors data and run approved workflows in Viably.
Modules 1 to 4
Engineering track
3 daysDevelopers, data engineers and platform teams
How the protocol works on the wire, then building, securing and deploying an MCP server for one of your own systems.
Modules 1, 2, 4 and 5
Bring MCP into your company with a team that builds it
Algoritma has trained staff at more than 500 companies across Indonesia and Singapore. Our parent company, Supertype, builds and runs the Sectors and Viably MCP servers, so the people teaching your team have shipped MCP in production. We scope each program with you first, and it fits alongside our AI adoption programs and corporate training. Individuals can join our MCP and AI agents bootcamp.
Your team finishes with
- AI clients connected to Sectors and Viably for every participant
- A playbook of which agent changes each role may approve
- An MCP security policy ready for review
- A working MCP server for one of your internal systems
- A 90 day rollout plan with owners and metrics