You re-explain the context every session
You open Claude, describe the project, the stack, the conventions. An hour later, new session — you start over. And again tomorrow.
OWASP security audits, code review, SQL optimization, E2E tests, Kubernetes, refactoring… Everything a senior dev would do — methodically, locally, on your own Claude subscription. One catalog, every editor.
›@security-auditor audit src/auth/
# Security audit — 2 findings
## CRITICAL login.php:42 — SQL injection. Fix: prepared statement.
## HIGH session.php:18 — Weak token (mt_rand). Fix: random_bytes(32).
›@sql-optimizer /orders query is slow
# Seq scan, 4.2M rows, missing composite index.
# Proposed: idx_orders(user_id, created_at DESC)
# Expected: 180ms → 4ms (45× faster)
›@commit-message-writer
Three frictions that cost you hours every week.
You open Claude, describe the project, the stack, the conventions. An hour later, new session — you start over. And again tomorrow.
You know there are bugs. Your eyes are tired. You skip the review. You ship. Monday-morning crash in prod.
“Later” means never. Tech debt piles up. Each feature is harder than the last.
The same 30 specialists, wherever you code.
Native Claude Code (CLI + Desktop)
/plugin marketplace add maifady/maifady-pluginView the plugin
Cursor rules (.mdc), auto-attached by file type
copy .cursor/rules/*View the Cursor rules
One npm package, every MCP client
npx -y maifady-mcpView the MCP server
30 agents that already know what to do, how to do it, and in which format.
Each agent has a precise methodology for its domain: OWASP Top 10 for the security auditor, EXPLAIN-ANALYZE for SQL, AAA for tests. No generic “figure it out”.
Add it with a single command (or copy a folder for Cursor). No sign-up, no API key to manage, no Maifady account. The MCP path just needs Node.js ≥ 18.
Agents run in your own editor, on your machine. No Maifady server. No telemetry. No ads. No data in transit.
Each category has its agents and their combinations.
Fast P0/P1/P2 review, deep PR review, OWASP audit, incremental refactoring, dead code, cognitive complexity.
EXPLAIN-driven SQL optimization, normalized schema design, safe zero-downtime migrations, latency/RAM/CPU profiling.
AAA unit tests, robust Playwright E2E, targeted high-impact coverage-gap detection.
Conventional Commits, README from the real repo, stack-aware .gitignore, ADR + runbooks, auto API docs.
React class → hooks migration, WCAG 2.2 AA audit, bundle analysis with ROI-ordered cuts.
Coherent REST/GraphQL design, OpenAPI 3.1 from code, owned micro-vs-monolith decisions.
Multi-stage Dockerfiles, prod-ready Kubernetes manifests (HPA, PDB, NetworkPolicy), multi-cloud Terraform.
LLM prompt improvement, production ML pipeline architecture with drift monitoring, legacy regex decoding.
Pick your editor. Zero account, no API key.
Run this in Claude Code, then restart your session:
/plugin marketplace add maifady/maifady-pluginOr git clone into ~/.claude/plugins/maifady/.
Plugin repo →
Copy the rules into your project (or your global Cursor config):
cp -r .cursor/rules/* your-project/.cursor/rules/Cursor auto-detects the rules — no restart needed.
Cursor repo →
Add to your MCP client config:
"command": "npx", "args": ["-y", "maifady-mcp"]Requires Node.js ≥ 18. One install, every MCP client.
MCP repo →
# Then, in any editor, just invoke an agent: ›@code-reviewer-lite review my staged changes ›@security-auditor audit auth/ against OWASP ›@sql-optimizer my /orders query is slow
Why 30 methodical agents beat one generic assistant — in every editor.
| Maifady | Custom GPTs | Cursor Agents | GitHub Copilot | |
|---|---|---|---|---|
| Price | 0 € | $20/mo | $20/mo | $10/mo |
| Open source MIT | ✓ | — | — | — |
| 100% local | ✓ | — | — | — |
| Specialized agents (≥10) | 30 | Manual | ~5 | — |
| Methodology per agent | ✓ | Write your own | Partial | — |
| Works in the terminal (Claude Code) | ✓ | — | — | — |
| Works in Cursor | ✓ rules + MCP | — | Native | Partial |
| Works in Claude Desktop / Cline / Zed | ✓ via MCP | — | — | — |
| Auditable / forkable | ✓ | — | — | — |
| Telemetry | None | Yes | Yes | Yes |
All answers are also indexed in JSON-LD for search engines and LLMs.
Yes, 100% free. MIT license, open code, no hidden cost, no freemium, no usage limit. You use your existing Claude (or other LLM) subscription — Maifady adds no cost and no intermediation.
No. Maifady is 100% local: no Maifady server, no telemetry, no analytics. The agents are just markdown files your editor reads locally. Model calls go through your own client — exactly like a normal conversation.
Three products cover them all: Maifady Plugin for Claude Code (CLI + Desktop), Maifady for Cursor as native .mdc rules, and Maifady MCP — one npm package that works in Cursor, Claude Desktop, Cline, Zed and any MCP client.
Use the plugin if you live in Claude Code. Use the Cursor rules for the tightest Cursor integration. Use the MCP server if you want a single install that works across Cursor, Claude Desktop, Cline and Zed at once — it's the most portable option.
No API key, ever — the agents run inside your editor's LLM, so there's no Maifady API and no extra quota or cost. Node.js ≥ 18 is only needed for the MCP server (the plugin and Cursor rules need nothing).
Drop any .md file into the plugin's agents/ folder. The format is simple: YAML frontmatter (name, description, optional tools) then the system prompt in markdown. Your editor loads them on next start.
By default, the model configured in your session (typically Sonnet 4.x). You can pin a specific model per agent via the frontmatter: model: opus, sonnet, haiku, or inherit.
None. No promo messages, no hidden monetization, no affiliate links. The project is fully open source MIT — you can audit every line.
Claude Code 1.0 and above. The agent format (YAML frontmatter + markdown system prompt) has been stable since Claude Code launched. Works on Windows, macOS and Linux.
Yes — MIT license, so you can fork, modify, redistribute, integrate into a commercial product. The only requirement: keep the MIT copyright and license notice.
Documentation and agents ship in 5 languages: English, French, Spanish, Chinese (Simplified) and Japanese.
Open an issue or PR on GitHub. For short questions, email hello@maifady.com.
30 agents. Every editor. 0 €. No account, no card, no setup.
Install Maifady