CLAUDE.md Template for Zephex Users

CLAUDE.md is the file Claude Code reads at session start. If you reference Zephex tools there, Claude Code can use them proactively instead of waiting for manual prompting.

THE TEMPLATE

Create a file called CLAUDE.md in your project root and paste this:

CLAUDE.md

# Project Instructions for Claude Code ## MCP Tools Available (via Zephex) You have access to Zephex MCP tools. Use them proactively: - **scope_task**: Call this FIRST before any non-trivial coding task. It returns the minimal files to read and existing utilities to reuse.- **get_project_context**: Call once per session for orientation. Do not call it repeatedly.- **read_code**: Use instead of reading full files when you only need a specific function or class.- **find_code**: Use to locate patterns, env vars, or symbols before assuming they don't exist.- **explain_architecture**: Call when making changes to auth, database access, or billing flows.- **audit_headers**: Call after every deploy to check security header configuration.- **thinking**: Call for complex multi-step debugging or high-risk changes to maintain session state. ## Tool Usage Rules 1. Always call scope_task before starting a coding task - it saves context2. Use read_code instead of reading whole files - pass the function/class name3. Use find_code before implementing anything new - it may already exist4. Call get_project_context at session start, not mid-session ## Project-Specific Notes - Runtime and package manager: Document the exact runtime, package manager, and build tool used here.- Key paths: List the directories Claude should inspect first for product code, tests, and configs.- Validation commands: Add the exact test, lint, and build commands Claude should run before finishing work.- Constraints: Record naming rules, architecture boundaries, deployment constraints, and anything Claude must avoid changing.

CUSTOMIZING IT

The Project-Specific Notes section is where you add your own stack details, key directories, naming conventions, deployment targets, and other repo-specific guidance. The tool instructions above it are generic and work across projects.

WHERE TO PUT IT

Project-level: ./CLAUDE.md in the repo root. This applies only to Claude Code sessions in that project.
Global: ~/.claude/CLAUDE.md. This applies to every Claude Code session on the machine.

Use the global path if you want Zephex tool guidance everywhere. Use the project root if you want more tailored instructions per repository.

FOR ALL OTHER EDITORS: AGENTS.MD

AGENTS.md is the cross-tool standard that works with OpenAI Codex, GitHub Copilot, Cursor, Windsurf, and most other coding agents. It uses the same concept as CLAUDE.md but works universally.

Create AGENTS.md at your project root:

AGENTS.md

# Zephex MCP Tools Available ## EndpointURL: https://zephex.dev/mcpAuth: Bearer YOUR_API_KEYProtocol: JSON-RPC 2.0 over HTTPS ## Tool Usage Rules1. Call scope_task FIRST before any coding task — it returns the minimal file set to read2. Call get_project_context once per session for orientation — not repeatedly3. Use read_code with a symbol name instead of reading full files4. Use find_code to search for patterns before assuming something doesn't exist5. Call thinking for complex multi-step debugging to maintain session state ## Available Tools- scope_task: Returns minimal files for a task (call this FIRST)- get_project_context: Project brief — stack, auth, hosting, key integrations, full file tree- read_code: AST-based symbol extraction with callers and imports- find_code: Fast repo-wide pattern search- explain_architecture: End-to-end request and auth flow tracing- check_package: Verify npm package exists, compare versions- audit_package: CVE check and migration guide for npm upgrades- audit_headers: HTTP/TLS security header audit- inspect_url: Extract clean content from any URL- thinking: Structured reasoning with persistent session state ## Project Notes[Add your project-specific context here]

Place it at ./AGENTS.md in your project root for most editors, or ~/.agents.md for global defaults.

WHICH FILE SHOULD YOU USE?

File	Works with
CLAUDE.md	Claude Code only
AGENTS.md	OpenAI Codex, GitHub Copilot, Cursor, Windsurf, JetBrains, and most others

For maximum coverage, create both files with the same content. Most codebases now have both.

EDITOR COMPATIBILITY

Editor	Reads AGENTS.md	Reads CLAUDE.md
OpenAI Codex	✓	✗
GitHub Copilot	✓	✗
Cursor	✓	✗
Windsurf	✓	✗
JetBrains AI	✓	✗
Claude Code	✗	✓

# Project Instructions for Claude Code ## MCP Tools Available (via Zephex) You have access to Zephex MCP tools. Use them proactively: - **scope_task**: Call this FIRST before any non-trivial coding task. It returns the minimal files to read and existing utilities to reuse.- **get_project_context**: Call once per session for orientation. Do not call it repeatedly.- **read_code**: Use instead of reading full files when you only need a specific function or class.- **find_code**: Use to locate patterns, env vars, or symbols before assuming they don't exist.- **explain_architecture**: Call when making changes to auth, database access, or billing flows.- **audit_headers**: Call after every deploy to check security header configuration.- **thinking**: Call for complex multi-step debugging or high-risk changes to maintain session state. ## Tool Usage Rules 1. Always call scope_task before starting a coding task - it saves context2. Use read_code instead of reading whole files - pass the function/class name3. Use find_code before implementing anything new - it may already exist4. Call get_project_context at session start, not mid-session ## Project-Specific Notes - Runtime and package manager: Document the exact runtime, package manager, and build tool used here.- Key paths: List the directories Claude should inspect first for product code, tests, and configs.- Validation commands: Add the exact test, lint, and build commands Claude should run before finishing work.- Constraints: Record naming rules, architecture boundaries, deployment constraints, and anything Claude must avoid changing.

# Zephex MCP Tools Available ## EndpointURL: https://zephex.dev/mcpAuth: Bearer YOUR_API_KEYProtocol: JSON-RPC 2.0 over HTTPS ## Tool Usage Rules1. Call scope_task FIRST before any coding task — it returns the minimal file set to read2. Call get_project_context once per session for orientation — not repeatedly3. Use read_code with a symbol name instead of reading full files4. Use find_code to search for patterns before assuming something doesn't exist5. Call thinking for complex multi-step debugging to maintain session state ## Available Tools- scope_task: Returns minimal files for a task (call this FIRST)- get_project_context: Project brief — stack, auth, hosting, key integrations, full file tree- read_code: AST-based symbol extraction with callers and imports- find_code: Fast repo-wide pattern search- explain_architecture: End-to-end request and auth flow tracing- check_package: Verify npm package exists, compare versions- audit_package: CVE check and migration guide for npm upgrades- audit_headers: HTTP/TLS security header audit- inspect_url: Extract clean content from any URL- thinking: Structured reasoning with persistent session state ## Project Notes[Add your project-specific context here]

File

Works with

CLAUDE.md

Claude Code only

AGENTS.md

OpenAI Codex, GitHub Copilot, Cursor, Windsurf, JetBrains, and most others

Editor

Reads AGENTS.md

Reads CLAUDE.md

OpenAI Codex

✓

✗

GitHub Copilot

✓

✗

Cursor

✓

✗

Windsurf

✓

✗

JetBrains AI

✓

✗

Claude Code

✗

✓