How it works

From codebase to deep context in minutes

Scope maps your entire codebase — entities, relationships, conventions, dependencies — and delivers it to any AI agent via MCP. Your agent starts every session already knowing your architecture.

Get started free Explore features

Step 1

Sync your codebase

Connect a GitHub repo or sync local files via scope_sync. In about two minutes, Scope extracts structured metadata that persists across sessions and powers context delivery.

What Scope extracts

Entities

User, Order, Product, Payment

Endpoints

POST /orders, GET /orders/:id, PATCH /payments/:id

Relationships

Orders belong to Users, contain Line Items

Tech Stack

TypeScript, Express, PostgreSQL

Conventions

Validation rules, naming patterns, architecture boundaries

How it works

Parses project structure, frameworks, and service boundaries

Reads schema and migration files to map entities and constraints

Builds endpoint + dependency graphs across files and modules

Indexes structured context for MCP delivery and feature spec generation

Supported: AST + schema parsing across 7+ languages, including Ruby, Python, TypeScript, JavaScript, Go, and Rust.

Step 2

Your AI agent gets structured context via MCP

Claude Code, Cursor, or any MCP client connects to Scope and pulls structured codebase context on demand. No copy-pasting, no stale docs.

What your agent can query

get_context

Pull entities, relationships, endpoints, and tech stack by scope

search

Semantic search across project patterns, decisions, and conventions

analyze

Run dependency and impact analysis before making changes

start_ticket

Get the next ticket with full implementation context

What this means in practice

Your agent knows every entity, relationship, and endpoint before writing a line of code

It follows your naming conventions and architecture patterns automatically

Cross-file dependencies are visible, so changes don't break things silently

Context persists across sessions. No re-exploration on every conversation

Tokens go to writing code, not reading files

Your agent can query other projects too — pull backend entities while working in the frontend

Step 3

Describe features, get codebase-grounded specs

Describe what you want to build in plain English. Scope maps your request to your actual code and generates implementation-ready tickets with files, dependencies, and acceptance criteria.

claude code

What Scope does

Maps your request to existing entities, endpoints, and services

Identifies exact files and modules that need to change

Flags missing capabilities your codebase doesn't have yet

Splits large features into properly scoped tickets

Orders dependencies so foundational work ships first

Generates testable acceptance criteria for each ticket

Step 4

Your agent builds complete features

With structured metadata delivered via MCP, your AI agent uses fewer tokens, follows your conventions, respects dependencies, and handles the parts it would normally miss.

MCP-powered executionPrimary

Claude Code, Cursor, and other MCP clients pull context automatically as they work

Agent calls get_context for entities, relationships, and conventions

Agent calls search to find relevant patterns and prior decisions

Agent calls start_ticket to get implementation specs with exact files

Agent calls complete_ticket to log learnings for future sessions

Also works with teams

Export tickets as Markdown or CSV for Jira, Linear, or any tracker

Specs include exact file paths, schema changes, and endpoint impact

Acceptance criteria and test steps are explicit and verifiable

Dependencies are ordered so teams know what ships first