Integrations
Use Delta with your favorite AI coding tools.
Claude Code (MCP Server — stdio)
Delta includes a built-in MCP (Model Context Protocol) server that integrates directly with Claude Code.
Setup
Add to .claude/settings.json in your project:
{
"mcpServers": {
"delta": {
"command": "npx",
"args": ["delta-ctx", "mcp"],
"description": "Delta Context Engine - optimized context per task"
}
}
}
How it works
- Claude Code detects the MCP server configuration
- Before each task, Claude calls Delta’s MCP tools
- Delta assembles the optimized context payload
- Claude receives only the relevant files, symbols, and memories
- 85% fewer tokens per task — automatically
Available MCP Tools (14)
| Tool | Description |
|---|---|
get_optimized_context | Get token-optimized context for a task |
get_community_map | Architectural community map |
get_execution_flows | Traced execution flows from entry points |
get_blast_radius | Impact analysis for changed files |
get_risk_scores | File risk scores (5 dimensions) |
get_memory | Query the persistent memory system |
save_memory | Save architectural knowledge |
search_codebase | Hybrid FTS5 + semantic search |
get_graph_diff | Architectural diff vs. a snapshot |
get_hub_files | Most architecturally central files |
get_bridge_files | Cross-community bridge connectors |
get_snapshot | List or retrieve graph snapshots |
save_snapshot | Save a new graph snapshot |
get_stats | Index statistics overview |
Available MCP Prompts (5)
| Prompt | Description |
|---|---|
blast_radius | “What does this change break?” — full impact analysis |
codebase_compass | “How is this codebase structured?” — architecture overview |
fault_tracer | “Trace this error to root cause” — debug assistant |
first_day | “Get a new dev productive in minutes” — onboarding guide |
merge_guardian | “Is this PR safe to merge?” — APPROVE / REVIEW / BLOCK |
Once configured, Claude Code uses Delta automatically. You don’t need to do anything differently — just chat with Claude as usual.
HTTP MCP Server (Universal)
Start Delta as an HTTP server so any MCP-compatible client can connect — not just Claude Code.
Setup
delta serve
This starts an HTTP MCP server at http://127.0.0.1:7734 with all 14 tools and 5 prompts.
Options
| Option | Default | Description |
|---|---|---|
--port <port> | 7734 | Server port |
--host <host> | 127.0.0.1 | Server host |
--root <path> | . | Project root directory |
Endpoints
| Endpoint | Method | Description |
|---|---|---|
/mcp | POST | JSON-RPC MCP requests |
/health | GET | Server status and tool count |
Example: Connect from any MCP client
# Start the server
delta serve --port 7734
# Health check
curl http://127.0.0.1:7734/health
# Call a tool via JSON-RPC
curl -X POST http://127.0.0.1:7734/mcp \
-H 'Content-Type: application/json' \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "search_codebase",
"arguments": { "query": "authentication" }
}
}'
OpenAI-Compatible Proxy
Delta can sit between any AI tool and your LLM provider, auto-injecting context into every request.
Setup
delta proxy --provider openai --api-key $OPENAI_API_KEY
How it works
- Your AI tool sends requests to
http://127.0.0.1:7735(the proxy) - Delta intercepts each request and extracts the task from the last user message
- Delta runs its full pipeline and injects optimized context as a system message
- The request is forwarded to your LLM provider with context included
- Zero configuration needed in your AI tool — just point it at the proxy URL
Supported Providers
| Provider | Flag | Default Model |
|---|---|---|
| OpenAI | --provider openai | gpt-4o |
| OpenAI Codex | --provider codex | o3-mini |
| Anthropic | --provider anthropic | claude-sonnet-4-20250514 |
| Google Gemini | --provider gemini | gemini-2.5-flash |
| OpenCode | --provider opencode | default |
| Local LLM | --provider local | llama3 (Ollama) |
Options
| Option | Default | Description |
|---|---|---|
--port <port> | 7735 | Proxy port |
--provider <name> | openai | LLM provider |
--model <model> | Provider default | Model name override |
--api-key <key> | $OPENAI_API_KEY | API key |
--root <path> | . | Project root |
Example: Use with any tool
# Start the proxy
delta proxy --provider openai
# Point your tool at the proxy
export OPENAI_API_BASE=http://127.0.0.1:7735/v1
# Now use your tool as normal — Delta context is auto-injected
your-ai-tool "fix the auth bug"
Cursor
Setup
delta cursor-init
This command:
- Creates
.cursor/ruleswith Delta-optimized instructions - Configures auto-update of
.delta/cursor-context.md
How it works
Every time you run delta run, the context is also written to .delta/cursor-context.md. Cursor reads this file as additional context for its AI completions.
Manual usage
You can also pipe Delta output directly:
# Generate context and let Cursor pick it up
delta run "your task" > .delta/cursor-context.md
VS Code Extension
Install
Search for “∆ Delta Context Engine” in the VS Code Marketplace, or:
code --install-extension delta-context-engine
Features
| Feature | Description |
|---|---|
| Sidebar | Live token savings per task |
| Context Manifest | Visual breakdown of which files are included and why |
| Status Bar | Token counter showing savings in real-time |
| Commands | One-click init, run, and watch from the command palette |
Available Commands
Open the command palette (Cmd+Shift+P / Ctrl+Shift+P) and type “Delta”:
Delta: Initialize ProjectDelta: Run TaskDelta: Watch ModeDelta: Show StatsDelta: Show Report
Any AI Assistant (CLI)
Delta outputs plain text to stdout, so it works with any AI assistant:
Copy to Clipboard
# macOS
delta run "your task" | pbcopy
# Linux (requires xclip)
delta run "your task" | xclip -selection clipboard
# Windows (PowerShell)
delta run "your task" | Set-Clipboard
Save to File
delta run "your task" > context.md
Pipe to Another Tool
# Use with any CLI tool that accepts stdin
delta run "your task" | your-ai-tool --context -
CI/CD Integration
GitHub Actions
Use Delta in your CI pipeline to generate architectural reports:
name: Delta Analysis
on: [push]
jobs:
analyze:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: '20'
- name: Install Delta
run: npm install -g delta-ctx
- name: Initialize
run: delta init
- name: Risk Report
run: delta risk
- name: Blast Radius (changed files)
run: |
for file in $(git diff --name-only HEAD~1); do
echo "=== $file ==="
delta blast "$file" 2>/dev/null || true
done
Pre-commit Hook
Add Delta analysis to your git pre-commit hook:
#!/bin/sh
# .git/hooks/pre-commit
echo "∆ Running Delta analysis..."
# Check for high-risk changes
for file in $(git diff --cached --name-only); do
risk=$(delta blast "$file" 2>/dev/null | grep "Risk:" | head -1)
if echo "$risk" | grep -q "HIGH"; then
echo "⚠ HIGH risk change detected: $file"
echo "$risk"
echo "Consider reviewing before committing."
fi
done
Architecture
Your Editor
│
├──── Claude Code ──── MCP Server (stdio) ──── Delta Core
│
├──── Any MCP Client ──── HTTP Server (:7734) ──── Delta Core
│
├──── Any AI Tool ──── Proxy (:7735) ──── LLM Provider
│ │
│ Delta auto-injects
│ context into requests
│
├──── Cursor ──── .cursor/rules + .delta/cursor-context.md
│
├──── VS Code ──── Extension ──── Delta CLI
│
└──── Any Tool ──── stdout / file ──── Delta CLI
│
┌──────┴──────┐
│ Delta Core │
│ │
│ • Change │
│ Detection │
│ • AST │
│ Parsing │
│ • Graph │
│ • Search │
│ • Embeddings│
│ • Memory │
│ • Assembly │
└──────────────┘
│
.delta/delta.db
(SQLite · local)