Merlin

The One AI Coding Agent CompanionYou'll Ever Need

Scroll
AI Coding Infrastructure

The AI coding infrastructureyour AI is missing.

34 agents. 43 commands. 20 hooks. One install.Merlin is the complete operating system for AI-assisted development.

Works with Claude Code, Codex CLI, OpenCode, Gemini CLIFree tier includes everything30-second install
See what's included

AI coding has a ceiling.
You've already hit it.

Your AI is brilliant — for 20 minutes. Then it forgets the auth flow. Creates a utility that already exists. Ignores your naming conventions. Can't plan past the next function.

You've been working around these limits. Building context files. Re-explaining architecture. Copy-pasting between sessions.

That's not AI-assisted development. That's AI-managed development. You're the manager.

you, every session
you:"No, the auth middleware is in /src/middleware/auth.ts"
you:"We already have a date formatter in utils/"
you:"I told you yesterday — we use camelCase here"
you:"Stop creating duplicate helpers"
you:"...let me re-explain the architecture again"
// 45 minutes spent managing your AI instead of building
Foundation

Your entire codebase.
Understood instantly.

Other AI tools read files one at a time and hope for the best. Merlin builds a complete knowledge graph of your entire codebase.

Every file's purpose. Every function's role. Every pattern and convention. How modules connect. What depends on what. Where things live and why.

And it never goes stale. Push to main? The knowledge graph updates automatically.

50 files or 50,000 — same speed
Auto-syncs on every push
Patterns & conventions extracted
Cross-file relationships mapped
Queryable by any AI tool
Anti-patterns documented
merlin sights
> merlin context "add OAuth authentication"
Querying knowledge graph...
Relevant files:
src/middleware/auth.ts — Existing auth middleware
src/routes/auth.ts — Auth route patterns
src/lib/session.ts — Session management
Patterns to follow:
Routes use requireAuth middleware
Error format: { error, code, message }
Tests in __tests__/ co-located
Anti-patterns:
Never inline auth checks — use middleware
Context ready in 47ms — no file scanning needed
1000+

Agents. Curated. Graded.
Always ready.

34 specialist agents ship built-in. 1000+ more are crawled daily from GitHub, npm, and MCP registries — reviewed, graded from A++ to B, and available instantly. You describe the task. Merlin picks the best agent.

34
built-in
1000+
cloud-indexed
Daily
crawled
A++
graded
Built-in specialists8 agents
Spec WriterTurns ideas into clear, buildable requirements
ArchitectDesigns minimal, clean architecture
BuilderWrites production code matching your patterns
CleanerFinds duplication, consolidates utilities
Security LeadValidation, error handling, auth checks
QA EngineerTest strategy for critical paths
DevOps LeadDeployment, env vars, build configs
Technical WriterDocumentation that stays current

We integrate the best so you don't have to chase it

The AI ecosystem moves fast — new agents, skills, and MCP servers appear daily. Merlin monitors all of it, grades quality, and makes the best tools available automatically. Built to complement Claude Code and other AI agents' evolution, never conflict. As your tools grow, Merlin grows with them.

GitHub repositoriesnpm packagesskills.sh marketplaceCommunity contributionsMCP server registriesSmithery.ai catalog
merlin smart route
> "Add Stripe webhook handling"
Searching index (1,247 agents)...
A++stripe-webhook-expert97% match
A+payment-integration89% match
Aapi-integrator82% match
Routing to stripe-webhook-expert — fresh 1M context

You describe what you want. Merlin decides who handles it. Each agent gets full codebase context. Each agent remembers your preferences.

Like a real engineering organization — except it's instant, tireless, and gets smarter every time you use it.

Idea → Spec → Architecture → Code → Security → Tests → Docs → Ship

The full engineering pipeline. Every step handled by the right specialist. Every step verified before moving on.

01
Define

"Build user authentication with OAuth"

02
Specproduct-spec

Requirements, flows, edge cases

03
Architectsystem-architect

Design approach, check conflicts

04
Implementimplementation-dev

Production code, reuse patterns

05
Hardenhardening-guard

Validation, error handling, security

06
Testtests-qa

Tests for critical paths

07
Verifyautomatic

Build, types, tests, security — all green

08
Documentdocs-keeper

Update relevant documentation

09
Ship

Production-ready. Not a prototype.

Trigger any step individually, or let Merlin run the full pipeline.
The system adapts to the size of the task.

Autonomous

Go to sleep.
Wake up to shipped features.

The Merlin Loop runs Claude autonomously for hours. Fresh 1M context every iteration. Crash recovery. Checkpoint persistence.

This isn't "auto-complete on steroids." This is real autonomous development:

→You write a plan with 30 tasks
→Run merlin-loop build
→Each task gets a fresh Claude with full context
→Builds, tests, commits — task by task
→Crashes? Recovers from checkpoint
→5 failures? Circuit breaker stops safely
1M
Context per task
Auto
Recovery
Circuit breaker
Safety
merlin-loop build — auth-system
Phase: User Authentication · 7/9 tasks complete
✓Set up database schema
✓Create user model & migrations
✓Add auth middleware
✓Build login/register endpoints
✓Add session management
✓Write auth tests
●Harden input validationbuilding...
○Add rate limiting
○Update API documentation
Runtime: 2h 14m0 errors · 14 commits

The full picture.

Everything that ships when you run one command.

34
specialist agents
43
slash commands
20
automated hooks
1000+
indexed from cloud
12
MCP tools
4
multi-agent workflows
4
supported runtimes
10
hook templates
3
loop modes

Three steps.

From install to production code.

01

Install

One command. Everything is included — agents, commands, hooks, loop.

npx create-merlin-brain
02

Describe

Tell Claude what you want to build. Merlin creates a plan and routes to the right agents.

/merlin:new-project
03

Ship

Production-ready code with tests, security hardening, and documentation.

git push

Works wherever
you code.

Merlin installs globally and exposes a standard MCP server. Any tool that supports MCP gets full Merlin capabilities.

Claude Code is the primary runtime — full feature parity. Other runtimes get the same context, agents, and memory.

Claude Code
primary
Codex CLI
OpenAI
OpenCode
open source
Gemini CLI
Google
Cursor
via MCP
Windsurf
via MCP
Any MCP tool
if it speaks MCP, Merlin works

Free. Actually free.

The full Merlin system — every agent, every command, the loop — ships free. No tricks. Sights is the optional cloud layer.

Free tier
$0

Forever. No credit card. No account.

  • 34 specialist agents
  • 43 slash commands
  • 20 automated hooks
  • 4 multi-agent workflows
  • Merlin Loop (3 modes)
  • Behavior learning
  • Project rules & memory
  • Full planning lifecycle
  • Quality verification
  • Code review & security
  • Multi-agent coordination
  • AFK mode with safety limits
  • All platform integrations
  • No account required
$ npx create-merlin-brain
Merlin Sights
$20
/month

The cloud layer. Cancel anytime.

Everything in free, plus:

  • Living wiki of your codebase
  • Cloud knowledge graph
  • Vector search (meaning-based)
  • Auto-sync on every push
  • Cross-session cloud memory
  • Public Sights catalog
  • Team collaboration
  • Cloud agent index
See Sights pricing

Your AI is working at 10%.

Fix that in 30 seconds.

Free forever. No credit card. No signup. Just install and go.