Onboard to a new codebase in one focused session

Joining a team or auditing an acquisition? Agents without MCP hallucinate folder layout and invent entry points. With Zephex they pull real structure first — MCP, CLI, or web terminal.

Problem

Day-one agent chats burn tokens grepping blindly, then write “docs” that describe a repo that does not exist. You need grounded stack + architecture + symbols before any edit.

Before / after

Before: “Looks like a Next.js app with Prisma…” (model guessed from file names).
After: Real scripts, auth method, deploy target, and ranked files for the task — agent opens the right files first.

Copy-paste agent prompt

Onboard me to this repo. Do not invent architecture.
1. get_project_context topic=identity
2. get_project_context topic=run then topic=auth
3. explain_architecture (overview)
4. find_code for the main HTTP entry and auth middleware
5. Summarize: stack, how to run tests, first 5 files I should read
Stop if a tool fails — do not fill gaps from training data.

CLI (no editor required)

cd /path/to/repo
zephex overview
zephex deep "onboard me" --json   # agent packet schema_version: 1
zephex architecture

Tool chain

get_project_context — stack, scripts, deps, env
explain_architecture — layers / Mermaid
find_code — jump to auth, DB, API routes
read_code — outline/symbol when paths are known

Full recipes · All solutions · Recommended MCP stack · Quickstart

Copy-paste agent prompt

Onboard me to this repo. Do not invent architecture. 1. get_project_context topic=identity 2. get_project_context topic=run then topic=auth 3. explain_architecture (overview) 4. find_code for the main HTTP entry and auth middleware 5. Summarize: stack, how to run tests, first 5 files I should read Stop if a tool fails — do not fill gaps from training data.

Onboard to a new codebase in one focused session

Joining a team or auditing an acquisition? Agents without MCP hallucinate folder layout and invent entry points. With Zephex they pull real structure first — MCP, CLI, or web terminal.

Problem

Day-one agent chats burn tokens grepping blindly, then write “docs” that describe a repo that does not exist. You need grounded stack + architecture + symbols before any edit.

Before / after

Before: “Looks like a Next.js app with Prisma…” (model guessed from file names).
After: Real scripts, auth method, deploy target, and ranked files for the task — agent opens the right files first.

Copy-paste agent prompt

Onboard me to this repo. Do not invent architecture.
1. get_project_context topic=identity
2. get_project_context topic=run then topic=auth
3. explain_architecture (overview)
4. find_code for the main HTTP entry and auth middleware
5. Summarize: stack, how to run tests, first 5 files I should read
Stop if a tool fails — do not fill gaps from training data.

CLI (no editor required)

cd /path/to/repo
zephex overview
zephex deep "onboard me" --json   # agent packet schema_version: 1
zephex architecture

Tool chain

get_project_context — stack, scripts, deps, env
explain_architecture — layers / Mermaid
find_code — jump to auth, DB, API routes
read_code — outline/symbol when paths are known

Full recipes · All solutions · Recommended MCP stack · Quickstart