Generate Code with MCP

Once Cobre’s public documentation MCP is connected, your AI agent can read the live OpenAPI Specification and generate code, models, tests, tickets, and acceptance criteria that match the published API — instead of guessing from training data.

This page is for product managers and developers. Use it after you complete Build with AI.

Always refresh MCP before you generate anything. The spec is cached locally. If Cobre published a field, enum, or error change, a stale cache will produce wrong code.

Refresh MCP and reload the latest API specification.

What MCP is for (and what it is not)

Use MCP for	Use documentation pages for	Do not use MCP for
Paths, methods, required fields, enums, examples, error shapes	Product behaviour, operating hours, reconciliation narrative, certification	Pricing, KYC, per-account limits, credentials, or unpublished rails

Best practice: ask the agent to use MCP for the contract and llms.txt for the surrounding guide. See Connect Cobre's Documentation to AI.

Treat generated code as a draft. Review every field name, enum, and status against the API Explorer before you merge it. If the agent and docs.cobre.com disagree, trust the published documentation.

How to prompt

Work in agent mode (the agent must be able to call MCP tools). Start with the business goal, then lock the contract, then generate.

Refresh the spec

Description: Ask the agent to refresh MCP.

What to expect after the action: The agent re-downloads the OpenAPI spec. Subsequent answers use current fields and examples.

Name the outcome

Description: Say what you need (a service class, a webhook handler, a Jira ticket, acceptance criteria) and the language or tracker you use.

What to expect after the action: The agent knows the deliverable before it opens schemas.

Ground in MCP

Description: Instruct the agent to read endpoints and schemas from MCP. Tell it not to invent fields, statuses, or error codes.

What to expect after the action: Generated types and payloads match the OAS, including required vs optional fields.

Review, then iterate

Description: Compare the output with the matching API Guide and Explorer page. Ask for a diff against the spec if anything looks invented.

What to expect after the action: A patch you can share with engineering, not a second guess.

Copy a prompt from the groups below and replace the placeholders (<language>, <path in your repo>, <geography>).

Product managers

Use these when you need to turn a product idea into something engineering can build this sprint — without waiting for a Solution Design if you only need a slice.

Map a requirement to endpoints

Draft acceptance criteria from the spec

Compare two geographies or products

Define the event model for a feature

Check a field, status, or error before committing

Produce an engineering kickoff brief

Developers

Use these in the repository where the integration will live. Name your language and the file path you want the agent to write.

Generate typed models from OAS schemas

Scaffold an authenticated HTTP client

Create a Counterparty, then a Money Movement

Build a webhook listener

Handle errors from the spec

Keep existing DTOs in sync with the OAS

Generate tests from official examples

Map reconciliation from webhooks and reports

Combine MCP with docs and demo flows

Source	When
MCP (OAS)	Fields, enums, examples, HTTP codes — any generated code or DTO
llms.txt + API Guides	Sequence, prerequisites, operating model — any design or ticket
Demo Flow Library	Teaching the sequence to a new engineer — Explore API Workflow Examples

Example prompt that uses all three:

Refresh MCP. Read https://docs.cobre.com/llms.txt for the payout guides. Then, using only OAS schemas, generate a <language> service that authenticates, creates a Counterparty, creates a Money Movement, and handles completion and failure webhooks. After the code, list the demo flow in https://cobre-demo-flows.cobre.co/index.html that is closest to this sequence. Do not invent fields.

Review checklist

Before you merge generated code or send a PM brief to engineering:

MCP was refreshed in this session

Every JSON key exists in the OAS (search the schema, do not rely on the chat)

Enums and status.state values are copied from the spec, not normalized into your own vocabulary

Webhook header and envelope fields match Notifications & Subscriptions

No pricing, KYC, or limit numbers unless they appear in the published docs

Open questions are listed, not filled in with guesses

When Cobre ships an API change, refresh MCP and re-run the Keep existing DTOs in sync prompt. See Managing API and Product Changes.

How to integrate with Cobre — client journey from pre-sales to go-live

Build with AI — connect the Cobre Docs MCP in your IDE

Create a Solution Design with AI — full integration design when the slice is larger than one ticket

Explore API Workflow Examples — interactive walkthroughs of the same sequences

Connect Cobre's Documentation to AI — use llms.txt for documentation-grounded answers

Generate Code with MCP

What MCP is for (and what it is not)#

How to prompt#

Product managers#

Developers#

Combine MCP with docs and demo flows#

Review checklist#

Related guides#

What MCP is for (and what it is not)

How to prompt

Product managers

Developers

Combine MCP with docs and demo flows

Review checklist

Related guides