에이전트 강화
게시자의 자체 URL입니다. PlugMCP 엔드포인트가 아닙니다.
Host a copy with our keys and logs.
https://api.machinegrade.dev/mcp위의 Claude 버튼을 누르면 이름과 URL이 채워진 상태로 Claude가 열립니다. 확인은 Claude에서 직접 해야 합니다.
claude mcp add --transport http io.github.machinegrade-validate https://api.machinegrade.dev/mcpRun in a terminal.
{ "mcpServers": { "io.github.machinegrade-validate": { "command": "npx", "args": [ "-y", "mcp-remote", "https://api.machinegrade.dev/mcp" ] } }}claude_desktop_config.json
원격 URL인가요, 내 컴퓨터에서 실행하는 명령인가요? 두 설치 방식의 차이
PlugMCP에서 실행
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
실행게시자 제공
Validate AI-generated artifacts against a contract before you act on them:
json_schema — validate artifact against a JSON Schema (all errors collected).openapi_response — validate a response body against the response schema for a given path + + in an OpenAPI spec.Deterministic validation for AI-generated artifacts: JSON Schema, OpenAPI response, SQL syntax.
methodstatussql — check a SQL string for syntax errors in a given dialect.Every check returns a verdict, not an error: {valid, errors, latency_ms},
HTTP 200 whether the artifact is valid or not. Only genuinely wrong requests
(bad key, unsupported type, malformed body, over your limit) get typed HTTP
errors.
Live at https://api.machinegrade.dev — free tier, self-service key, try it in 30 seconds (first example below). Built on Hono; the same codebase runs on Cloudflare Workers (production) and plain Node (local dev), and is MIT-licensed if you'd rather self-host.
Agents that generate JSON, API responses, or SQL need a fast, cheap, machine-checkable pass/fail before they ship the result — cheaper than a full LLM-as-judge call, and deterministic.
npm install
npm run dev
# machinegrade validate listening on http://localhost:8787
# Get an API key (live service — works as-is)
curl -s -X POST https://api.machinegrade.dev/keys \
-H 'content-type: application/json' \
-d '{"email": "[email protected]"}'
# => {"key":"sk_..."}
# Validate a JSON artifact against a JSON Schema
curl -s -X POST https://api.machinegrade.dev/v1/validate \
-H 'content-type: application/json' \
-H 'X-Api-Key: sk_...' \
-d '{
"type": "json_schema",
"artifact": {"name": "Ada", "age": 30},
"contract": {
"schema": {
"type": "object",
"required": ["name", "age"],
"properties": {"name": {"type": "string"}, "age": {"type": "number"}}
}
}
}'
# => {"valid":true,"errors":[],"latency_ms":1}
import requests
base = "http://localhost:8787"
key = requests.post(f"{base}/keys", json={"email": "[email protected]"}).json()["key"]
resp = requests.post(
f"{base}/v1/validate",
headers={"X-Api-Key": key},
json={
"type": "sql",
"artifact": "SELECT id, name FROM users WHERE id = 1",
"contract": {"dialect": "mysql"},
},
)
print(resp.status_code, resp.headers.get("X-Calls-Remaining"), resp.json())
mcp/server.ts exposes a single tool, validate, that forwards to
POST /v1/validate. Point an MCP-compatible client at it:
{
"mcpServers": {
"machinegrade-validate": {
"command": "npx",
"args": ["tsx", "mcp/server.ts"],
"cwd": "/path/to/validate",
"env": {
"SANDBOX_URL": "http://localhost:8787",
"SANDBOX_API_KEY": "sk_..."
}
}
}
}
The production service also exposes an MCP endpoint directly — no local process, no npm install — via streamable HTTP at:
POST https://api.machinegrade.dev/mcp
It's the same single validate tool as the stdio adapter above.
initialize and tools/list work without a key (discovery is
anonymous); tools/call requires X-Api-Key (issue one via POST /keys, same as the REST API — the free tier and limits are shared).
With Claude Code:
claude mcp add --transport http validate https://api.machinegrade.dev/mcp --header "X-Api-Key: sk_..."
The stdio adapter via npm (@machinegrade/validate, see above) remains
available for local/offline use or clients without HTTP transport
support.
Download the latest .mcpb bundle from
Releases and
double-click it. Claude Desktop asks for an API key during install and stores it
in the OS keychain (macOS Keychain, Windows Credential Manager) rather than in a
config file — nothing lands in claude_desktop_config.json.
Issue a free key (500 calls/month) first:
curl -s -X POST https://api.machinegrade.dev/keys \
-H 'content-type: application/json' \
-d '{"email": "[email protected]"}'
The bundle is a thin stdio client over the hosted API: ~34 KB, zero bundled dependencies. If you self-host, point the extension's API base URL setting at your own deployment and the tool talks to that instead.
See public/openapi.yaml for the full contract, or
/v1/manifest for a machine-readable
summary (types, limits, pricing, error codes) once the service is running.
/llms.txt is a short pointer for LLM agents.
POST /v1/paid-request (requires X-Api-Key);
you'll be notified when it's live.Every error is typed JSON — {code, message, hint, docs_url} — never a
free-form string:
A verdict ({valid, errors, latency_ms}) is never an error — an
invalid artifact is a normal, expected outcome and returns HTTP 200.
artifact must be a JSON value, not a JSON-encoded string:
{"type": "json_schema", "artifact": {"name": "Ada"}, "contract": {"schema": {"type": "object"}}} // correct
{"type": "json_schema", "artifact": "{\"name\": \"Ada\"}", "contract": {"schema": {"type": "object"}}} // wrong
For type: "sql" the artifact is a string — the statement itself.
Because MCP callers stringify values often enough (and did so through Claude
Desktop until the tool schema declared artifact's types), the service
tolerates the wrong form narrowly: if artifact is a string, the type is
json_schema or openapi_response, and the contract's top-level schema
declares types that exclude string, the string is JSON-decoded before
validation and the verdict carries "decoded_from_string": true. The field is
additive; {valid, errors, latency_ms} is unchanged.
It deliberately does not decode otherwise, because a string artifact is often legitimate:
src/storage.ts defines a Storage interface with two implementations:
MemoryStorage — full in-memory implementation, used for npm run dev
and the test suite.D1Storage — real Cloudflare D1 binding, backed by schema.sql (keys,
events tables). Used in production; the Workers entry point in
src/index.ts builds it from the DB binding on first request.Apply schema.sql to a new D1 database with:
wrangler d1 execute machinegrade-validate-db --file=schema.sql # local
wrangler d1 execute machinegrade-validate-db --file=schema.sql --remote # production
npm test # vitest run, in-process via app.request(), MemoryStorage
npm run typecheck # tsc --noEmit
Tests cover: key issuance, happy + fail cases for each validator, typed
401/400/402/429 errors, the metering limits (both injectable in tests so
they don't require looping hundreds of real requests), and /stats funnel
counts.
The service runs on Cloudflare Workers (Hono + D1 + Workers Static Assets). To self-host on a fresh Cloudflare account:
wrangler d1 create machinegrade-validate-db # copy the returned database_id into wrangler.toml
wrangler d1 execute machinegrade-validate-db --file=schema.sql --remote
wrangler secret put ADMIN_TOKEN
wrangler deploy
Then bind a custom domain (e.g. api.machinegrade.dev) to the Worker via
the Cloudflare dashboard or wrangler. CI can deploy on push to main once
CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID repo secrets are set and
the deploy job in .github/workflows/ci.yml is uncommented.
Two things worth knowing about the Workers port:
GET /openapi.yaml and GET /llms.txt are served by the ASSETS binding
([assets] in wrangler.toml, pointing at public/) — Cloudflare serves
them directly, without invoking the Worker. The routes in src/index.ts
are a fallback for local Node dev/tests, where there's no ASSETS binding.json_schema and openapi_response validators use
@cfworker/json-schema, not ajv: ajv compiles schemas via
new Function(...), which the Workers runtime disallows, and schemas
here arrive dynamically per request (from the caller), so they can't be
precompiled at build time either.Early stage, honestly so: this service is live and free-tier usage is real, and we're measuring whether it earns a paid tier. What you can rely on:
/v1/validate request/response shapes, typed error
codes, verdict semantics) is stable — breaking changes only with a
versioned path (/v2/...), never silently.Feedback and integration stories are the most valuable thing you can give
us right now: open an issue or use POST /v1/paid-request if you need
more than the free tier.
Deterministic validation for AI-generated artifacts: JSON Schema, OpenAPI response, SQL syntax.
{ "mcpServers": { "io.github.machinegrade-validate": { "url": "https://api.machinegrade.dev/mcp", "headers": {} } }}도구
아직 도구 목록이 없습니다. 원격 서버는 첫 상태 확인 후에 도구 목록이 생깁니다.
FAQ
이 페이지에서 Cursor, Claude Desktop, Claude Code 또는 VS Code용 http 설정을 복사합니다. PlugMCP 프록시가 아니라 게시자의 설치 방식을 사용합니다.
게시자가 등록한 환경 변수: SANDBOX_URL, SANDBOX_API_KEY.
Validate에는 HTTP로 연결합니다.
마지막 확인: 2026-11-09. Deterministic validation for AI-generated artifacts: JSON Schema, OpenAPI response, SQL syntax.