v1.4.9 — live on npm
One command crawls your entire codebase and gives Claude Code, Cursor, or Windsurf the context it needs to write code that actually fits your project.
free to use · byok · works with any repo
scroll
// demo
Run it once. Your agent never starts a session blind again.
agentbrain — your-project
// without agentbrain
✗
Wastes the first 10 minutes getting oriented
✗
Writes code that doesn't match your patterns
✗
Asks what files to look at for every task
✗
Hallucinates function names and import paths
✗
You re-explain the architecture every session
// with agentbrain
✓
Knows every module, service, and file by name
✓
Follows your exact coding patterns automatically
✓
Jumps straight to the right files for any task
✓
Context stays fresh — auto-updates on every commit
✓
Works across Claude Code, Cursor, and Windsurf
// how_it_works()
AgentBrain runs once and works invisibly in the background forever.
01
Install + run setup
One command installs AgentBrain and scans your entire repo. Paste your API key when prompted. Done.
02
Context is generated
AgentBrain crawls your codebase, summarises every relevant file, and synthesises a navigation guide your agent reads.
03
Just code normally
Your agent loads context at every session start. After each commit, context refreshes silently in the background.
// features
Built for developers who ship with AI agents every day.
smart_context_generation()
Crawls your repo, scores file relevance, and produces a navigation guide with exact file paths and export names — not generic descriptions.
auto_inject_into_agents()
Automatically writes context loading instructions into CLAUDE.md, .cursorrules, and .windsurfrules. Zero manual steps required.
watch_commits_refresh()
A smart git hook detects source file changes and regenerates context silently after every commit. Doc-only changes are skipped.
cache_by_git_hash()
Context is cached against your HEAD commit hash. Repeat runs with no code changes cost exactly $0.00 and complete instantly.
bring_your_own_key()
Uses your Anthropic or OpenAI API key directly. Auto-detects provider from key prefix. No markup. No middleman.
generate_handoff_doc()
Run agentbrain handoff at end of session. Reads your git diff and writes a handoff.md your agent reads at next session start.
// compatibility
Generates the right config file for each agent automatically.
Claude Code
→ CLAUDE.md
Cursor
→ .cursor/rules
Windsurf
→ .windsurfrules
$0.18
typical cost production repo
$0.00
repeat runs no file changes
183
files analysed 1,200 file repo
<2m
time to full codebase intel
// mcp_server
Connect AgentBrain as an MCP server and let your existing subscription do the compute. You pay nothing extra.
claude_desktop_config.json
{ "mcpServers": { "agentbrain": { "command": "npx", "args": ["-y", "@agentbrain/mcp-server"] } } }
Claude Desktop
~/Library/Application Support/ Claude/claude_desktop_config.json
Cursor
Cmd+Shift+P MCP: Configure Servers
Windsurf
Settings → MCP Servers Add Configuration
// tools available to your agent
load_context
Load full codebase context at session start
load_standards
Load coding standards mid-session
save_handoff
Save session handoff at end of session
scan_repo
Inspect repo structure — no API key needed
// pricing
AgentBrain is completely free right now. You pay only for your own API tokens. Team features coming soon.
Install AgentBrain, run setup, and your agent will never start a session blind again.