Usage Guide

Learn how to use the core features of Cortex MCP

1

Project Initialization

When using Cortex for the first time in a new project, ask the AI to initialize.

Example Usage

"Initialize Cortex for this project"

Choose Scan Mode

FULL Deep analysis of entire codebase (high token usage, recommended for large projects)
LIGHT Scan only core files (README, config files, quick start)
NONE Skip scanning (simple tasks, testing)
2

Saving Context

When you want to save important work, ask the AI to update the context.

Example Usage

"Save the work so far to the context"
"Update the memory"

Note: Context saving does not run automatically. Request it explicitly after important work.

3

Loading Context

When starting a new session, you can load previous work context.

Example Usage

"Load the previous context"
"Let's continue from yesterday's work"

Tip: Cortex uses saved summaries to quickly understand the core of previous work and continue.

4

Hallucination Verification Core

Cortex's most important feature

When AI says "I modified the file" or "I added the function", verify if it actually did by checking the codebase.

Example Usage

"Verify what you just said"
"Check if this is a hallucination"
"You said the code changes are done, verify if they actually are"

How It Works

1
Claim Extraction

Auto-extract verifiable claims from AI response

2
AST Code Analysis

Parse syntax to confirm function/class exists

3
Auto Git Evidence

Auto-detect staged/unstaged file changes

4
Grounding Score

Quantify reliability with 0.0-1.0 score

Interpreting Results

ACCEPT Grounding Score >= 0.6 - Sufficient evidence, trustworthy
WARN 0.4 - 0.6 - Partially unverifiable, manual check recommended
REJECT < 0.4 - Insufficient evidence, rework needed

Coding-Specific Verification

Files

Existence

Functions

Signature

Classes

Structure

Lines

Number accuracy

Tip - Auto Evidence Collection: When verifying after file modifications, evidence is automatically collected from Git diff. No need to specify file lists separately.

5

Context Branch Management

Not Git branches, but Cortex context branching

When switching to a different topic or task, you can create a context branch to separate conversation contexts. This is managed internally by Cortex, independent of Git branches.

Note: "Branch" here does NOT mean Git branch. It's a conversation context branch stored by Cortex. Example: Separate "Bug Fix" context from "New Feature" context.

Example Usage

"Create a new context branch - Bug Fix"
"This is a separate topic, please split the context"

Auto Detection: When the AI detects a topic change, it will automatically ask "Should I create a new context branch?"

6

Context Snapshots

Backup of conversation context, not code

Save context state at important moments as snapshots, and restore when needed. This is NOT code backup - it's a backup of conversation history and summaries stored in Cortex.

Note: Snapshots do NOT backup source code. They backup conversation context that Cortex remembers (previous work, summaries, branch info, etc.). Use Git for code backup.

Create Context Snapshot

"Take a context snapshot"

Restore Context Snapshot

"Restore to yesterday's context snapshot"

Key Summary

1. Initialize = Start project
2. Save context = Manual save after work
3. Load context = At session start
4. Verify = Confirm when in doubt
5. Context Branch = When switching topics
6. Context Snapshot = Conversation backup