@ruah-dev/conv
v0.4.0
Published
Standalone package for converting API specs into agent-ready tool surfaces.
Maintainers
Readme
ruah conv
Convert API specs into agent-ready tool surfaces.
Feed it an OpenAPI spec, get MCP tool definitions, function-calling schemas, host-ready plugin bundles, or a full MCP server scaffold. One intermediate representation in the middle — every input parser normalizes to it, every output generator reads from it.
Input Parsers IR Output Generators
───────────── ────────────── ─────────────────────
OpenAPI 3.x ──┐ ┌── MCP Tool Definitions ✓
Swagger 2.0 ──┤ ├── MCP Server (TS) ✓
Postman v2.1 ──┼──→ Ruah Tool Schema ───┼── MCP Server (Python) ✓
GraphQL SDL ──┤ (canonical IR) ├── Function Calling (OpenAI) ✓
HAR files ──┘ ├── Function Calling (Anthropic) ✓
├── Claude Code plugin bundle ✓
├── Codex plugin bundle ✓
└── A2A service wrapper ✓Now shipped: OpenAPI 3.x, Swagger 2.0, Postman v2.1, and GraphQL SDL inputs; MCP JSON defs; MCP TypeScript/Python scaffolds; OpenAI and Anthropic tool schemas; Claude Code and Codex plugin scaffolds; A2A wrapper scaffold; config-file support.
See It
# Parse a spec and see what's inside
npx @ruah-dev/conv inspect petstore.yamlAPI Spec Summary
──────────────────────────────────────────────────
Title: Petstore API
Version: 1.0.0
Format: openapi-3.0
Base URL: https://api.petstore.example.com/v1
Auth Schemes (1)
• apiKeyAuth: apiKey (X-API-Key)
Tools (4)
• listPets GET /pets (2 params)
• createPet POST /pets (0 params +body) [moderate]
• getPet GET /pets/{petId} (1 params)
• deletePet DELETE /pets/{petId} (1 params) [destructive]
Types (3)
Pet, NewPet, Error# Generate MCP tool definitions
npx @ruah-dev/conv generate petstore.yaml --json[
{
"name": "listPets",
"description": "List all pets",
"inputSchema": {
"type": "object",
"properties": {
"limit": { "type": "integer", "minimum": 1, "maximum": 100 },
"offset": { "type": "integer" }
}
}
},
{
"name": "createPet",
"description": "Create a new pet",
"inputSchema": {
"type": "object",
"properties": {
"name": { "type": "string" },
"tag": { "type": "string" }
},
"required": ["name"]
}
}
]Use It
# Generate MCP tool definitions (JSON)
ruah conv generate ./spec.yaml --target mcp-tool-defs --json
# Generate OpenAI tool/function schemas
ruah conv generate ./spec.yaml --target openai-tools --json
# Generate a TypeScript MCP server scaffold
ruah conv generate ./spec.yaml --target mcp-ts-server --output ./generated-server
# Force a safer or broader OpenAPI/Swagger operation preset
ruah conv generate ./spec.yaml --operation-profile read-only
ruah conv generate ./spec.yaml --operation-profile all
# Generate a Claude Code plugin bundle
ruah conv generate ./spec.yaml --target claude-code-plugin-ts --output ./generated-claude-plugin
# Generate a Codex plugin bundle
ruah conv generate ./spec.yaml --target codex-plugin-ts --output ./generated-codex-plugin
# Write output to a directory
ruah conv generate ./spec.yaml --output ./generated/
# Inspect a spec — see tools, types, auth at a glance
ruah conv inspect ./spec.yaml
# Validate a spec — check for issues before generating
ruah conv validate ./spec.yaml
# Full IR as JSON — for piping into other tools
ruah conv inspect ./spec.yaml --json
# List available output targets
ruah conv targetsOr use the standalone CLI directly:
npx @ruah-dev/conv generate ./spec.yaml --jsonHow It Works
1. Parse
The parser reads OpenAPI 3.0/3.1, Swagger 2.0, Postman v2.1 collections, and GraphQL SDL, then normalizes everything into the Ruah Tool Schema — the canonical intermediate representation.
ruah conv inspect ./spec.yaml --json | jq '.tools | length'
# 122. Validate
The IR is validated for completeness. Warnings are advisory — the IR is always usable, but you'll know what's missing.
ruah conv validate ./spec.yaml
# ✓ Spec is valid — 12 tools, 8 types, no warnings.| Warning | What it means |
|---------|--------------|
| missing-description | Tool has no description — LLMs need this |
| missing-base-url | No server URL — tools will need a runtime base URL |
| duplicate-tool-name | Two operations produced the same name |
| unknown-type | Schema couldn't be parsed — treated as any |
| unresolved-ref | $ref target not found in components |
| no-operations | Spec has no paths — nothing to convert |
3. Generate
The generator reads the IR and produces output for the target format.
For OpenAPI 3.x and Swagger 2.0 inputs, generate now asks which operation preset to expose unless you set --operation-profile, configure operationProfile, or run non-interactively. Non-interactive runs default to standard.
Current targets:
mcp-tool-defs— JSON array of MCP-compatible tool definitionsmcp-ts-server— TypeScript MCP server scaffold with stdio entrypoint and HTTP startermcp-python-server— FastMCP Python server scaffoldopenai-tools— OpenAI tool/function definitionsanthropic-tools— Anthropic tool definitionsclaude-code-plugin-ts— Claude Code plugin scaffold with a generated TypeScript MCP servercodex-plugin-ts— Codex plugin scaffold with a generated TypeScript MCP servera2a-wrapper— A2A-style service wrapper scaffold
Each tool gets:
- name — normalized from
operationId(or synthesized from method + path) - description — from the spec's
summaryordescription - inputSchema — JSON Schema combining path params, query params, and request body
Paginated operations are detected heuristically and annotated in generated descriptions. Current patterns: limit + offset, page-number/page-size, and cursor-style query params.
OpenAPI/Swagger operation presets:
read-only— includesGET,HEAD,OPTIONSstandard— includesread-onlyplusPOST,PUTall— includesstandardplusPATCH,DELETE
4. Naming Policy
Operations become tool names using this priority:
- Use
operationIdif present →listPets - Fall back to
method + path→getPetsPetId - Deduplicate with numeric suffix →
listPets,listPets2
Normalization: get_user_by_id → getUserById, api.v2.list-pets → listPets
5. Risk Classification
Every tool gets a risk level based on HTTP method:
| Method | Risk | Idempotent | Read-Only |
|--------|------|-----------|-----------|
| GET, HEAD, OPTIONS | safe | Yes | Yes |
| POST | moderate | No | No |
| PUT | moderate | Yes | No |
| PATCH | destructive | No | No |
| DELETE | destructive | Yes | No |
The IR: Ruah Tool Schema
The intermediate representation is the core of Convert. Everything normalizes to this, everything generates from this.
interface RuahToolSchema {
meta: SchemaMeta; // title, version, baseUrl, source format
auth: AuthSchema[]; // apiKey, http/bearer, oauth2
tools: Tool[]; // one per API operation
types: Record<string, TypeDefinition>; // named schemas
}The IR is inspectable: ruah conv inspect ./spec.yaml --json dumps the full IR. Useful for debugging, validating, and piping into other tools.
Adding a new input format = one parser that produces IR. Adding a new output format = one generator that consumes IR. Never N×M.
Programmatic API
import { parse, validateIR, generate } from "@ruah-dev/conv";
// Parse a spec
const ir = parse("./petstore.yaml");
// Validate
const warnings = validateIR(ir);
// Generate MCP tool definitions
const result = generate("mcp-tool-defs", ir);
console.log(result.files[0].content);CLI Reference
ruah conv generate <spec> [options] Parse spec and generate output
--target <id> Output target (default: mcp-tool-defs)
--output <dir> Output directory (default: stdout)
--name <value> Override generated server/service name
--transport <mode> Generator transport hint
--operation-profile <profile> OpenAPI/Swagger preset: read-only, standard, all
--config <path> Load generation defaults from JSON
--json Output as JSON
ruah conv inspect <spec> [options] Parse spec and display IR summary
--json Output full IR as JSON
ruah conv validate <spec> [options] Parse spec and report warnings
--json Output warnings as JSON
ruah conv targets [options] List available output targets
--json Output as JSON
--help, -h Show help
--version, -v Show versionConfig File
Generator defaults can live in ruah.conv.json, .ruah-conv.json, or ruah.conv.config.json:
{
"target": "mcp-ts-server",
"output": "./generated/server",
"name": "Billing MCP Server",
"transport": "streamable-http",
"operationProfile": "standard"
}If operationProfile is omitted, interactive OpenAPI/Swagger runs prompt for a preset. Non-interactive runs default to standard.
Roadmap
v0.1 — Core pipeline
- [x] OpenAPI 3.0 / 3.1 parser
- [x] Ruah Tool Schema IR
- [x] IR validation with warnings
- [x] MCP tool definitions (JSON)
- [x] CLI: generate, inspect, validate, targets
- [x] Plug-and-play with ruah CLI
v0.2 — Expand
- [x] Full MCP TypeScript server scaffold (stdio transport)
- [x] Swagger 2.0 auto-upgrade
- [x] Postman collection parser
- [x] HTTP/SSE transport for MCP servers
- [x] Pagination handling
v0.3 — Multi-target
- [x] MCP Python server (FastMCP)
- [x] Function calling schemas (OpenAI, Anthropic)
- [x] GraphQL SDL parser
- [x] A2A service wrapper
- [x] Config file support
- [x] Claude Code and Codex plugin bundle targets
Install
Stable release:
npm install -g @ruah-dev/cli
ruah conv generate ./spec.yaml --jsonStandalone package:
npm install -g @ruah-dev/conv
ruah-conv generate ./spec.yaml --json
# or run without installing
npx @ruah-dev/conv generate ./spec.yaml --jsonRequirements: Node.js 18+. Single runtime dependency: yaml.
Release Flow
Releases are tag-driven:
git push origin main
git push origin v0.3.0Pushing a v* tag triggers the GitHub Actions release pipeline, which:
- runs typecheck, lint, and tests for the root package and
packages/core - publishes
@ruah-dev/conv-corefirst - waits for that version to appear on npm
- publishes
@ruah-dev/conv - creates the GitHub release automatically
Ecosystem
ruah — top-level CLI router (@ruah-dev/cli)
conv — standalone conv package + CLI wrapper (@ruah-dev/conv)
conv-core — conversion implementation (@ruah-dev/conv-core)Community
License
MIT
