Agentify

Agent Interface Compiler — One command. Every agent speaks your product.

---

Agentify compiles any OpenAPI specification into 8 agent interface formats — MCP Server, CLAUDE.md, AGENTS.md, .cursorrules, Skills, llms.txt, GEMINI.md, and A2A Card. Instead of hand-building each format separately, generate them all from a single source of truth.

npx agentify-cli transform https://petstore.swagger.io/v2/swagger.json

The Problem

AI agents are the new users of your API. But making your product agent-accessible requires building and maintaining multiple interface formats:

Format	Who consumes it	Manual effort
MCP Server	Claude, ChatGPT, Copilot	Days of coding
CLAUDE.md	Claude Code	Write from scratch
AGENTS.md	Codex, Copilot, Cursor, Gemini CLI	Write from scratch
.cursorrules	Cursor IDE	Write from scratch
Skills	30+ agent platforms	Per-platform work
llms.txt	LLM search engines	Manual authoring
GEMINI.md	Gemini CLI	Write from scratch
A2A Card	Google Agent-to-Agent protocol	JSON schema work

That's 8+ formats to build, test, and keep in sync. Every API change means updating all of them.

The Solution

Agentify is a compiler. OpenAPI in, every agent format out.

                    +---> MCP Server (with Dockerfile)
                    |
                    +---> CLAUDE.md
                    |
                    +---> AGENTS.md
                    |
OpenAPI Spec -----> +---> .cursorrules
                    |
                    +---> Skills
                    |
                    +---> llms.txt
                    |
                    +---> GEMINI.md
                    |
                    +---> A2A Card

Quick Start

# Transform any OpenAPI spec
npx agentify-cli transform https://petstore.swagger.io/v2/swagger.json

# Specify output directory
npx agentify-cli transform ./my-api.yaml -o ./output

# Override project name
npx agentify-cli transform https://api.example.com/openapi.json -n my-project

# Generate only specific formats
npx agentify-cli transform ./my-api.yaml -f mcp claude.md

Output:

  Agentify v0.3.0
  Agent Interface Compiler

  +-- 20 endpoints detected -> SMALL API strategy
  +-- 3 domains identified (pet, store, user)
  +-- Auth: apiKey (SWAGGER_PETSTORE_API_KEY)
  +-- Strategy: Direct tool mapping — one tool per endpoint

  > Generated mcp + claude.md + agents.md + cursorrules + llms.txt + gemini.md + skills + a2a (15 files)
  > Output: ./swagger-petstore-mcp-server
  > Security scan: PASSED

Features

Smart Strategy Selection — Automatically chooses the right generation strategy based on API size:

API Size	Endpoints	Strategy	Why
Small	< 30	Direct mapping	One tool per endpoint, simple and complete
Medium	30-100	Direct mapping (Tool Search planned)	Detects scale; optimized generation coming soon
Large	100+	Direct mapping (Code Exec planned)	Detects scale; context-optimized generation coming soon

Security First — Every generated artifact passes through:

• Input sanitization (blocks eval, exec, Function constructor, require/import injection)
• Handlebars template injection prevention
• Prompt injection pattern detection
• Generated code security scanning

Production Ready — Generated MCP servers include:

• TypeScript source with full type safety
• Dockerfile for containerized deployment
• Environment variable configuration (.env.example)
• Stdio transport (standard MCP protocol)

Output Format Status

Format	Status	Description
MCP Server	Available	Full server with tools, handlers, Dockerfile
CLAUDE.md	Available	Project context for Claude Code
AGENTS.md	Available	Universal agent instructions (Linux Foundation standard)
.cursorrules	Available	Cursor IDE agent rules
Skills	Available	Structured capability file for agent platforms
llms.txt	Available	LLM-readable condensed documentation
GEMINI.md	Available	Gemini CLI project context
A2A Card	Available	Google Agent-to-Agent discovery card

How It Works

1. PARSE        OpenAPI 3.x / Swagger 2.0 spec (URL or file)
                  |
2. SANITIZE     Strip dangerous patterns from all spec fields
                  |
3. ANALYZE      Detect domains, auth, scale -> pick strategy
                  |
4. COMPILE      Generate AgentifyIR (intermediate representation)
                  |
5. EMIT         Run selected emitters (MCP, Skills, Docs, etc.)
                  |
6. SCAN         Security scan all generated code
                  |
7. OUTPUT       Write files to disk

AgentifyIR is the canonical intermediate representation — a flat, typed structure that captures everything an emitter needs: product metadata, capabilities (endpoints), domains, auth config, and generation strategy.

Architecture

agentify/
+-- src/
|   +-- cli.ts              # CLI entry point (Commander.js)
|   +-- parser/             # OpenAPI parsing + input sanitization
|   +-- generator/          # Pluggable emitters for each format
|   |   +-- templates/      # Handlebars templates
|   +-- security/           # Input sanitization + output scanning
|   +-- types.ts            # AgentifyIR type definitions
+-- templates/              # Generated project templates
+-- test/                   # Vitest test suite

Contributing

Agentify welcomes contributions, especially new emitters (output formats). Each emitter implements a simple interface:

import type { Emitter, AgentifyIR, EmitterOptions, EmitterResult } from "../types";

export class MyFormatEmitter implements Emitter {
  readonly name = "my-format";
  readonly format = "my-format";

  async emit(ir: AgentifyIR, options: EmitterOptions): Promise<EmitterResult> {
    // Generate output files from the IR
    return { filesWritten: [...], warnings: [] };
  }
}

See CONTRIBUTING.md for the full guide.

Roadmap

• M0: Foundation — OpenAPI parser, MCP emitter, security scanner, CLI
• M1: Multi-Format — CLAUDE.md, AGENTS.md, Skills, .cursorrules, llms.txt, GEMINI.md, A2A Card
• M2: Intelligence — Capability graph, semantic grouping, context optimization
• M3: Self-Serve — Web UI, one-click deploy, registry integrations
• M4: Scale — Enterprise features, custom emitters, CI/CD integration

Compared to Alternatives

Feature	Agentify	Speakeasy	Stainless	openapi-to-skills
MCP Server	Yes	Yes	No	No
Skills	Yes	CLI only	No	Yes
CLAUDE.md	Yes	No	No	No
AGENTS.md	Yes	No	No	No
.cursorrules	Yes	No	No	No
llms.txt	Yes	Yes	No	No
GEMINI.md	Yes	No	No	No
A2A Card	Yes	No	No	No
Context-aware strategy	Yes	No	Yes	No
Security scanning	Yes	Unknown	Unknown	No
Open source	MIT	No	No	MIT

No existing tool compiles one OpenAPI spec into all agent interface formats.

License

MIT -- Agentify Contributors