Skip to content

OpenHelm via MCP & API

A virtual computer your agent calls for the work it can't do alone.

OpenHelm spins up a secure remote environment (a real stealth browser, a coding agent, an email inbox and more) to execute one-off or scheduled tasks on behalf of your AI. Access it via zero-install MCP servers in ChatGPT, Claude or Cursor, or call the REST API from your backend.

Connect an MCP serverRead the API docs →

Use-case MCP servers

Purpose-built environments for the tasks agents actually need to execute.

Developers, engineering teams & power users

General-Purpose

Spin up a virtual computer for any custom task, browse, run code, read data, and return structured results.

mcp.openhelm.ai/main/mcp

Investment analysts, VCs & corporate strategy teams

Deep Research

A virtual research environment that reads SEC filings, earnings prints, and live market data, grounded in official sources.

mcp.openhelm.ai/research/mcp

Growth marketers, content directors & SEO specialists

SEO & GEO

A virtual SEO environment: audit rankings, draft content, and measure AI engine citations of your brand.

mcp.openhelm.ai/seo/mcp

Founders, sales & ops teams, and AI agents

Email Outreach

A virtual inbox your agent controls, send, monitor replies, and run the whole conversation to completion.

mcp.openhelm.ai/email/mcp

Engineering teams, indie developers & AI coding agents

Dev Changes

A virtual coding environment: clone the repo, make the change, run tests, and open a PR.

mcp.openhelm.ai/dev/mcp

Product, marketing & competitive-intelligence teams

Site Monitoring

Watch websites, competitors and prices; get evidence-backed diffs and briefs instead of noisy pixel alerts.

mcp.openhelm.ai/watch/mcp

Coming soon

Soon

Phone Reach-Out

A virtual phone your agent uses to call, converse, and act on the outcome, coming soon.

Soon

WhatsApp Reach-Out

A virtual WhatsApp inbox your agent uses to message, handle replies, and achieve the goal, coming soon.

Soon

Online Buying

A virtual shopper that sources products, compares options, and checks out within your guardrails, coming soon.

Two ways in

Your assistant gets the virtual computer. Or build it into your product.

  • Remote MCP servers

    Add a URL in ChatGPT, Claude or Cursor. No install, no key handling. Your assistant gets tools that spin up a full virtual computer in the cloud to do the actual work.

  • REST API (/v1)

    POST /v1/runs with a prompt + optional JSON output_schema. Projects, jobs, runs, connections and email, server-to-server from your own backend.

  • Stealth browser, deeper web access

    The virtual computer drives a browser that is undetectable by anti-bot systems, giving the agent access to more of the web than standard automation tools.

  • Goals achieved, not just tasks dispatched

    A monitoring layer watches work in progress and ensures the objective is actually reached, not just that a tool call returned a response.

  • Multi-step orchestration

    Complex tasks are broken into sequential steps managed by an agentic system that adapts when things change, rather than a single prompt-to-output call.

  • Real budgets

    Per-key monthly caps, an org daily cap, and a cost estimate on every started run. Usage draws from your plan's credits.

REST API

Or call it directly from your backend.

POST /v1/runs
curl https://api.openhelm.ai/v1/runs \
  -H "Authorization: Bearer oh_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Get the pricing tiers from stripe.com",
    "output_schema": { "type": "object",
      "properties": { "tiers": { "type": "array", "items": { "type": "string" } } } }
  }'
202 + poll/webhook
HTTP/1.1 202 Accepted
{
  "task_id": "run_2x9f4c...",
  "poll_url": "https://api.openhelm.ai/v1/runs/run_2x9f4c...",
  "billing": { "model": "sonnet", "estimated_cost_usd": { "low": 0.05, "high": 0.3 } }
}

# poll until done (or supply callback_url for a webhook):
curl https://api.openhelm.ai/v1/runs/run_2x9f4c... -H "Authorization: Bearer oh_live_..."

Full endpoints, the result envelope, webhooks, and the Connections API are in the developer docs, and the live API reference is generated from the deployed OpenAPI spec.

Get started via MCP

Connect in under a minute.

  1. Pick a server

    Choose from the purpose-built servers above, or start with the General-Purpose server for custom prompts and full API access from your assistant.

  2. Add the remote URL to your assistant

    In ChatGPT, Claude, Cursor or any MCP client, add the server URL as a remote connector. No install, no terminal, no Node setup required.

  3. Connect your OpenHelm account

    Click Connect and authorise via OAuth 2.1. Your encrypted OpenHelm API key is wrapped in a signed token; the server itself stores no secrets.

  4. Run tasks from your assistant

    Ask your assistant in plain English. OpenHelm spins up the virtual computer (stealth browser, code runner, or email environment), executes the task, and returns evidence-backed structured results.

Common questions

What is the difference between the MCP servers and the REST API?

Same engine, two front doors. The MCP servers are the zero-install way: add a URL in ChatGPT, Claude or Cursor, connect once via OAuth, and your assistant can run tasks directly. The REST API (/v1) is for building OpenHelm into your own product or backend, server-to-server. Both run work in isolated cloud sandboxes and return the same structured, evidence-backed results.

How do responses work for tasks that take minutes, hours, or run forever?

Every run returns a task_id. Fast tasks resolve inline; for longer ones you poll GET /v1/runs/{id} or supply a callback_url to receive an HMAC-signed webhook when it completes. Recurring or scheduled jobs fire that webhook on every run, so a long-lived monitor can stream results back for weeks. The MCP servers use a bounded wait then a check_result continuation so connections never hang.

How does the agent get the credentials it needs?

Connections. Create token (paste-a-key) or remote MCP connections directly over POST /v1/connections, or mint a hosted link for OAuth providers (Google, GitHub, Notion…). Runs then resolve whatever connections the task needs. A run that needs a connection that is not set up fails with failure_reason "auth_required" rather than guessing.

How is it priced?

Usage draws from your OpenHelm plan's monthly credits (1 credit = $0.05 of billed cost). Each started run returns a cost estimate, and the actual cost comes back in usage.cost_usd when it completes. Per-key monthly caps and an org daily cap bound spend. New accounts add a card to start a free trial.

Is my key safe in a public MCP listing?

Yes. The MCP servers are stateless and store no secrets. Connecting wraps your encrypted OpenHelm key in a signed OAuth token. Card-required trials protect the shared cloud compute from multi-account abuse.

Give your agent a virtual computer. Zero-install.

Connect a remote MCP server in ChatGPT, Claude or Cursor in under a minute, with a stealth browser, coding environment, and email inbox ready to go. Or mint an API key and build OpenHelm directly into your own product.

Connect an MCP serverGet an API key →