Convertrilo

Official registry Productivity

Create and manage Convertrilo video encoding jobs from MCP clients and agent workflows.

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

NPM package @convertrilo/sdk
Claude Code
claude mcp add convertrilo -- npx -y @convertrilo/sdk@0.1.2
JSON configuration (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "convertrilo": {
            "command": "npx",
            "args": [
                "-y",
                "@convertrilo/sdk@0.1.2"
            ]
        }
    }
}

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Contents
  1. Convertrilo TypeScript SDK
  2. VMAF And Encoding Passes
  3. Install
  4. CLI And MCP Automation
  5. Create A Client
  6. Examples
  7. Idempotent Job Creation
  8. Job File Cleanup
  9. Saved Encode Presets
  10. Saved Output Destinations
  11. URL Source To CDN Output
  12. URL Source To S3 Output
  13. URL Source To Google Drive Output
  14. Folder Ingest
  15. Webhook Delivery History
  16. Regenerate Types

Convertrilo TypeScript SDK

Type-safe client for the Convertrilo API.

VMAF And Encoding Passes

When optimize: "vmaf" is requested, Convertrilo performs VMAF sampling and one optimized final encode. A supplied passes: 2 value remains accepted for backward compatibility, but the API uses and bills one effective pass.

Pricing responses expose requestedPasses and effectivePasses. New integrations should use passes: 1 with VMAF.

Install

pnpm add @convertrilo/sdk

The package currently targets modern Node.js runtimes with global fetch. If your runtime does not provide fetch, pass fetchImpl to the client.

CLI And MCP Automation

The package includes two executable automation tools:

export CONVERTRILO_API_KEY="cvr_..."

convertrilo encode https://example.com/input.mp4 \
  --codec h264 \
  --resolution 1080p \
  --audio-policy transcode-aac \
  --frame-rate-policy cap \
  --scale-policy no-upscale \
  --quality better \
  --wait \
  --json

convertrilo-mcp

Use convertrilo for scripts, CI, cron, and local operations. Use convertrilo-mcp as a stdio MCP server for agent and workflow integrations.

Run convertrilo login or convertrilo init to save an API key locally, or run convertrilo with no arguments to start the interactive encode wizard. Waiting commands show terminal progress by default and keep --json clean for scripts.

Shell completion snippets are available with:

convertrilo completion zsh

See docs/CLI-AND-MCP.md for commands, MCP client config, S3 output examples, VMAF, two-pass, status, cancel, and balance workflows.

If you want the simplest terminal walkthrough, start with docs/CLI-QUICKSTART.md.

For agent/client setup, use docs/MCP-QUICKSTART.md.

MCP registry metadata lives in server.json.

Create A Client

import { ConvertriloClient } from "@convertrilo/sdk";

const client = new ConvertriloClient({
  baseUrl: "https://api.convertrilo.com",
  apiKey: process.env.CONVERTRILO_API_KEY,
});

Use an API key for server-to-server integrations. Browser apps should call your own backend, then your backend calls Convertrilo.

Examples

The examples/ directory contains starter scripts for the main integration paths:

  • node-url-to-cdn.ts - encode a public URL and receive a signed CDN download URL
  • node-url-to-s3.ts - encode a public URL and upload the result to S3/S3-compatible storage
  • google-drive-byo-token.ts - upload output to Google Drive using a customer-owned service account
  • folder-ingest-s3.ts - queue one encode job per video in an S3 prefix
  • idempotency.ts - safely retry createJob and createJobsBulk
  • webhook-receiver-hmac.ts - verify managed webhook HMAC signatures from a Node receiver

Local SDK smoke tests use .env, but published SDK users should provide credentials through their own server environment. Do not put Convertrilo API keys or customer storage tokens in frontend code.

For a complete server-to-server walkthrough covering URL, S3, folder ingest, Google Drive service accounts, polling, and webhooks, see docs/API-INTEGRATION-GUIDE.md.

Idempotent Job Creation

Use an idempotency key when retrying create calls from your backend. Reusing the same key with the same body returns the original response instead of creating duplicate jobs.

const job = await client.createJob({
  externalId: "upload-123",
  metadata: { customerId: "cus_123" },
  codec: "h264",
  resolution: "1080p",
  fps: 30,
}, {
  idempotencyKey: "job-upload-123",
});

const batch = await client.createJobsBulk({
  jobs: [
    {
      externalId: "batch-42:clip-1",
      codec: "h264",
      resolution: "1080p",
      fps: 30,
      sourceS3: { bucket: "source", key: "clip-1.mp4" },
    },
  ],
  settings: { confirm: true },
}, {
  idempotencyKey: "bulk-batch-42",
});

Job File Cleanup

Delete managed upload and output objects after a job reaches a terminal state:

await client.cancelJob(jobId); // required first for created, queued, or running jobs

const result = await client.deleteJobFiles(jobId);
console.log(result.objectsDeleted, result.jobRetained);

The job record and billing history remain available for auditing. Active jobs return the stable job_active error until they are canceled.

Saved Encode Presets

Save reusable encode settings without storing source URLs, storage credentials, or output secrets:

const preset = await client.createEncodePreset({
  name: "Default 1080p web MP4",
  settings: {
    codec: "h264",
    resolution: "1080p",
    fps: 30,
    preset: "standard",
    bitrateTier: "medium",
    passes: 1,
    policy: "fastest",
    container: "mp4",
    quality: "better",
    optimize: "none",
    vmafTarget: 93,
    audioPolicy: "transcode-aac",
    frameRatePolicy: "cap",
    scalePolicy: "no-upscale",
  },
});

const { presets } = await client.getEncodePresets();

Saved Output Destinations

Save reusable delivery targets without storing raw storage secrets in the destination itself. S3 destinations reference an encrypted saved S3 credential:

const destination = await client.createOutputDestination({
  name: "Customer uploads bucket",
  config: {
    type: "s3",
    credentialId: "0f5a7f2b-4ff2-45d4-b76f-1f7b6e98d4d1",
    keyPrefix: "processed/",
  },
});

const { destinations } = await client.getOutputDestinations();

URL Source To CDN Output

const job = await client.onDemandEncode({
  sourceUrl: "https://example.com/input.mp4",
  externalId: "customer-video-123",
  metadata: {
    customerId: "cus_123",
    workflow: "daily-compression",
  },
  codec: "h264",
  resolution: "1080p",
  quality: "better",
  audioPolicy: "transcode-aac",
  frameRatePolicy: "cap",
  scalePolicy: "no-upscale",
}, {
  idempotencyKey: "encode-customer-video-123",
});

let finalStatus;
while (true) {
  finalStatus = await client.onDemandStatus(job.jobId);

  if (finalStatus.status === "success") break;
  if (finalStatus.status === "failed") {
    throw new Error(finalStatus.failureMessage || "Encoding failed");
  }

  await new Promise((resolve) => setTimeout(resolve, 5000));
}

console.log(finalStatus.downloadUrl);
console.log(finalStatus.requestedExecution);
console.log(finalStatus.effectiveExecution);
console.log(finalStatus.sourceProbe?.color);
console.log(finalStatus.outputProbe);

Terminal users can inspect the same report with:

convertrilo status JOB_ID --json
convertrilo wait JOB_ID --json

URL Source To S3 Output

const job = await client.onDemandEncode({
  sourceUrl: "https://example.com/input.mp4",
  codec: "h264",
  resolution: "1080p",
  outputS3: {
    bucket: "customer-output-bucket",
    key: "encoded/input-1080p.mp4",
    region: "us-east-1",
    accessKeyId: process.env.CUSTOMER_S3_ACCESS_KEY_ID,
    secretAccessKey: process.env.CUSTOMER_S3_SECRET_ACCESS_KEY,
  },
});

console.log(job.jobId);

For S3-compatible services, pass endpoint and usually forcePathStyle: true.

URL Source To Google Drive Output

For headless API integrations, save a customer-owned Google service account once. Use a Google Shared Drive for output and add the returned service-account email as a member with permission to create files.

const credential = await client.createGoogleDriveCredential({
  name: "Production Drive",
  serviceAccount: JSON.parse(process.env.GOOGLE_SERVICE_ACCOUNT_JSON!),
});

const job = await client.onDemandEncode({
  sourceUrl: "https://example.com/input.mp4",
  codec: "h264",
  resolution: "1080p",
  outputGoogleDrive: {
    folderId: "GOOGLE_DRIVE_FOLDER_ID",
    fileName: "input-1080p.mp4",
    credentialId: credential.id,
  },
});

console.log(job.jobId);

Dashboard Google Picker authorization is separate from SDK automation.

Folder Ingest

Queue one job per video in an S3 prefix:

const batch = await client.onDemandIngestFolder({
  externalIdPrefix: "batch-2026-06-09",
  metadata: {
    customerId: "cus_123",
    workflow: "folder-compression",
  },
  sourceS3: {
    bucket: "customer-source-bucket",
    prefix: "incoming/",
    region: "us-east-1",
    accessKeyId: process.env.CUSTOMER_S3_ACCESS_KEY_ID,
    secretAccessKey: process.env.CUSTOMER_S3_SECRET_ACCESS_KEY,
  },
  outputDestination: "s3",
  outputS3: {
    bucket: "customer-output-bucket",
    prefix: "encoded/",
    region: "us-east-1",
    accessKeyId: process.env.CUSTOMER_S3_ACCESS_KEY_ID,
    secretAccessKey: process.env.CUSTOMER_S3_SECRET_ACCESS_KEY,
  },
  codec: "h264",
  maxFiles: 25,
  resolution: "1080p",
}, {
  idempotencyKey: "folder-batch-2026-06-09",
});

for (const job of batch.jobs || []) {
  console.log(job.jobId, job.externalId, job.fileName);
}

Use maxFiles to cap how many discovered videos are queued from a folder.

Queue one job per video in a Google Drive folder:

const batch = await client.onDemandIngestFolder({
  sourceGoogleDrive: {
    folderId: "SOURCE_FOLDER_ID",
    credentialId: credential.id,
  },
  outputDestination: "google-drive",
  outputGoogleDrive: {
    folderId: "OUTPUT_FOLDER_ID",
    credentialId: credential.id,
  },
  codec: "h264",
  maxFiles: 25,
  resolution: "1080p",
});

for (const job of batch.jobs || []) {
  console.log(job.jobId, job.fileName);
}

Convertrilo mints short-lived Google tokens from the encrypted service-account credential when each worker starts.

Poll each returned jobId with client.onDemandStatus(jobId).

Webhook Delivery History

Managed webhooks are HMAC signed. You can test a webhook and inspect recent delivery attempts:

await client.testWebhook(webhookId);

const history = await client.getWebhookDeliveries(webhookId);
for (const delivery of history.deliveries || []) {
  console.log(delivery.status, delivery.statusCode, delivery.event);
}

Regenerate Types

The SDK types are generated from openapi.yaml.

pnpm run generate
pnpm run build

The generate script uses --default-non-nullable false so OpenAPI defaults remain optional in TypeScript request bodies.

Frequently asked questions

What is Convertrilo?

Create and manage Convertrilo video encoding jobs from MCP clients and agent workflows.

How do I install Convertrilo?

It runs on your machine. In Claude Code: claude mcp add convertrilo -- npx -y @convertrilo/sdk@0.1.2. For other clients, paste the JSON configuration from this page into their MCP settings.

Does Convertrilo need an API key?

Its registry entry lists no required key or secret. Some features may still need an account with the service it connects to.

Is Convertrilo open source?

Yes. Its code is on GitHub under the MIT license.

Who maintains Convertrilo?

It is published by serkandrgn. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.