Generate GraphQL Schemas with Claude (2026)
The Problem
Designing a GraphQL schema requires careful thought about types, queries, mutations, input types, pagination, error handling, and naming conventions. Getting the schema wrong early leads to painful breaking changes later because GraphQL clients depend on exact field names and types.
Quick Start
Describe your data model and ask Claude Code to generate the schema:
Generate a GraphQL schema for a task management API.
Entities: User, Project, Task, Comment.
Include:
- Queries with pagination (cursor-based)
- Mutations for CRUD operations
- Input types with validation descriptions
- Proper nullability (use ! where values are guaranteed)
- Relay-style connection types for lists
Use the code-first approach with TypeGraphQL and TypeScript.
What’s Happening
GraphQL schemas define the contract between client and server. Unlike REST, where the API shape is implicit, GraphQL schemas are explicit and self-documenting. Every query, mutation, type, and field must be declared in the schema.
Claude Code excels at schema generation because it understands GraphQL conventions (Relay connections, input types, error unions), can derive the schema from your existing database models, and produces consistent naming throughout.
Step-by-Step Guide
Step 1: Choose your approach
Schema-first (write the schema, generate code):
# schema.graphql
type User {
id: ID!
email: String!
name: String!
projects: [Project!]!
createdAt: DateTime!
}
Code-first (write TypeScript, generate schema):
// Using TypeGraphQL
@ObjectType()
class User {
@Field(() => ID)
id: string;
@Field()
email: string;
@Field()
name: string;
@Field(() => [Project])
projects: Project[];
}
Ask Claude Code which approach fits your project:
I have an existing Prisma schema with 12 models. Should I use
schema-first or code-first for my GraphQL API? I want type safety
between the database models and the GraphQL types.
Step 2: Generate types from your data model
Ask Claude Code to read your Prisma schema (or database models) and generate GraphQL types:
Read prisma/schema.prisma and generate GraphQL type definitions
for every model. Follow these conventions:
- Use Relay-style connections for list fields
- Make IDs non-nullable
- Include timestamps (createdAt, updatedAt)
- Exclude internal fields (passwordHash, deletedAt)
- Use proper GraphQL scalar types (DateTime, JSON)
Claude Code generates:
scalar DateTime
type User {
id: ID!
email: String!
name: String!
role: UserRole!
projects: ProjectConnection!
tasks(status: TaskStatus): TaskConnection!
createdAt: DateTime!
updatedAt: DateTime!
}
enum UserRole {
ADMIN
MEMBER
VIEWER
}
type Project {
id: ID!
name: String!
description: String
owner: User!
tasks(first: Int, after: String, status: TaskStatus): TaskConnection!
members: [User!]!
createdAt: DateTime!
updatedAt: DateTime!
}
type Task {
id: ID!
title: String!
description: String
status: TaskStatus!
priority: TaskPriority!
assignee: User
project: Project!
comments(first: Int, after: String): CommentConnection!
dueDate: DateTime
createdAt: DateTime!
updatedAt: DateTime!
}
enum TaskStatus {
TODO
IN_PROGRESS
IN_REVIEW
DONE
}
enum TaskPriority {
LOW
MEDIUM
HIGH
URGENT
}
type Comment {
id: ID!
body: String!
author: User!
task: Task!
createdAt: DateTime!
updatedAt: DateTime!
}
Step 3: Generate connection types for pagination
Relay-style pagination is the GraphQL standard:
type TaskConnection {
edges: [TaskEdge!]!
pageInfo: PageInfo!
totalCount: Int!
}
type TaskEdge {
node: Task!
cursor: String!
}
type PageInfo {
hasNextPage: Boolean!
hasPreviousPage: Boolean!
startCursor: String
endCursor: String
}
Ask Claude Code to generate these for every list field:
Generate Relay-style connection types for all list relationships
in the schema. Include totalCount on every connection.
Step 4: Generate queries and mutations
type Query {
# Single resource lookups
user(id: ID!): User
project(id: ID!): Project
task(id: ID!): Task
# List queries with pagination and filtering
users(first: Int, after: String, search: String): UserConnection!
projects(first: Int, after: String, ownerId: ID): ProjectConnection!
tasks(
first: Int
after: String
projectId: ID
status: TaskStatus
assigneeId: ID
priority: TaskPriority
): TaskConnection!
# Current user
me: User!
}
type Mutation {
# User mutations
updateProfile(input: UpdateProfileInput!): User!
# Project mutations
createProject(input: CreateProjectInput!): Project!
updateProject(id: ID!, input: UpdateProjectInput!): Project!
deleteProject(id: ID!): Boolean!
addProjectMember(projectId: ID!, userId: ID!, role: UserRole!): Project!
# Task mutations
createTask(input: CreateTaskInput!): Task!
updateTask(id: ID!, input: UpdateTaskInput!): Task!
deleteTask(id: ID!): Boolean!
assignTask(taskId: ID!, userId: ID): Task!
updateTaskStatus(taskId: ID!, status: TaskStatus!): Task!
# Comment mutations
addComment(input: AddCommentInput!): Comment!
deleteComment(id: ID!): Boolean!
}
Step 5: Generate input types
input CreateProjectInput {
name: String!
description: String
}
input UpdateProjectInput {
name: String
description: String
}
input CreateTaskInput {
projectId: ID!
title: String!
description: String
priority: TaskPriority = MEDIUM
assigneeId: ID
dueDate: DateTime
}
input UpdateTaskInput {
title: String
description: String
priority: TaskPriority
assigneeId: ID
dueDate: DateTime
}
input AddCommentInput {
taskId: ID!
body: String!
}
input UpdateProfileInput {
name: String
email: String
}
Step 6: Generate resolvers
Ask Claude Code to generate type-safe resolvers:
Generate Fastify/Mercurius resolvers for the GraphQL schema.
Use the Prisma client for data access. Include:
- DataLoader for N+1 prevention
- Authorization checks
- Input validation
- Error handling
// src/graphql/resolvers/task.ts
import { GraphQLContext } from '../context';
import DataLoader from 'dataloader';
export const taskResolvers = {
Query: {
task: async (_: unknown, { id }: { id: string }, ctx: GraphQLContext) => {
ctx.requireAuth();
const task = await ctx.prisma.task.findUnique({ where: { id } });
if (!task) throw new Error('Task not found');
return task;
},
tasks: async (_: unknown, args: TasksArgs, ctx: GraphQLContext) => {
ctx.requireAuth();
const { first = 20, after, projectId, status, assigneeId } = args;
const where = {
...(projectId && { projectId }),
...(status && { status }),
...(assigneeId && { assigneeId }),
deletedAt: null,
};
const cursor = after ? { id: after } : undefined;
const tasks = await ctx.prisma.task.findMany({
where,
take: first + 1,
cursor,
skip: cursor ? 1 : 0,
orderBy: { createdAt: 'desc' },
});
const hasNextPage = tasks.length > first;
const edges = tasks.slice(0, first).map((task) => ({
node: task,
cursor: task.id,
}));
return {
edges,
pageInfo: {
hasNextPage,
hasPreviousPage: !!after,
startCursor: edges[0]?.cursor ?? null,
endCursor: edges[edges.length - 1]?.cursor ?? null,
},
totalCount: await ctx.prisma.task.count({ where }),
};
},
},
Task: {
assignee: (task: { assigneeId: string | null }, _: unknown, ctx: GraphQLContext) => {
if (!task.assigneeId) return null;
return ctx.loaders.user.load(task.assigneeId);
},
project: (task: { projectId: string }, _: unknown, ctx: GraphQLContext) => {
return ctx.loaders.project.load(task.projectId);
},
comments: async (task: { id: string }, args: PaginationArgs, ctx: GraphQLContext) => {
// Paginated sub-query
const { first = 20, after } = args;
const comments = await ctx.prisma.comment.findMany({
where: { taskId: task.id },
take: first + 1,
cursor: after ? { id: after } : undefined,
skip: after ? 1 : 0,
orderBy: { createdAt: 'asc' },
});
const hasNextPage = comments.length > first;
const edges = comments.slice(0, first).map((c) => ({
node: c,
cursor: c.id,
}));
return {
edges,
pageInfo: { hasNextPage, hasPreviousPage: !!after },
totalCount: await ctx.prisma.comment.count({ where: { taskId: task.id } }),
};
},
},
};
Step 7: Generate DataLoaders for N+1 prevention
Generate DataLoader instances for all relationships that could
cause N+1 queries. Register them in the GraphQL context so they
are created fresh per request.
// src/graphql/loaders.ts
import DataLoader from 'dataloader';
import { PrismaClient } from '@prisma/client';
export function createLoaders(prisma: PrismaClient) {
return {
user: new DataLoader<string, User>(async (ids) => {
const users = await prisma.user.findMany({
where: { id: { in: [...ids] } },
});
const userMap = new Map(users.map((u) => [u.id, u]));
return ids.map((id) => userMap.get(id) ?? new Error(`User ${id} not found`));
}),
project: new DataLoader<string, Project>(async (ids) => {
const projects = await prisma.project.findMany({
where: { id: { in: [...ids] } },
});
const projectMap = new Map(projects.map((p) => [p.id, p]));
return ids.map((id) => projectMap.get(id) ?? new Error(`Project ${id} not found`));
}),
};
}
Prevention
Add GraphQL conventions to your CLAUDE.md:
## GraphQL Rules
- Use Relay-style connections for all list fields
- Every mutation input must be a dedicated Input type
- Use enums for finite value sets
- Non-nullable (!) by default, nullable only when data is absent
- Include totalCount on all connections
- Use DataLoader for all relationship resolvers
- Never expose internal IDs or sensitive fields
I'm a solo developer in Vietnam. 50K Chrome extension users. $500K+ on Upwork. 5 Claude Max subscriptions running agent fleets in parallel.
These are my actual CLAUDE.md templates, orchestration configs, and prompts. Not a course. Not theory. The files I copy into every project before I write a line of code.
**[See what's inside →](https://zovo.one/lifetime?utm_source=ccg&utm_medium=cta-default&utm_campaign=claude-code-graphql-schema-generation-guide)**
$99 once. Free forever. 47/500 founding spots left.
Related Guides
- Claude Code Database Schema Design Guide
- Claude Code API Endpoint Testing Guide
- Claude Code API Contract Testing Guide