AlgoHandYour assistant already talks. Let it act.
Your customers' EAs run on VPS machines your assistant cannot reach. AlgoHand is the missing layer: one API that starts, stops and reconfigures Expert Advisors on those machines — with the validation, confirmation and proof a regulated business needs.
POST /v1/accounts/acc_9q1x7d/commands
{ "client_id": "ea-cust-1042",
"type": "set_param",
"params": { "instance_id": "inst_g7h8i9",
"inputs": { "MaximumSpread": 15 } },
"dry_run": true }
200 summary: "MaximumSpread 7 → 15 on Night Hunter Pro
(EURUSD) — EA will restart"
diff: { MaximumSpread: { was: 7, now: 15 } }
restarts_ea: true · open_positions: 0
… the customer says yes …
202 command_id: cmd_01M1JY85A07NCXYZ
GET /v1/commands/cmd_…?wait=30
200 status: succeeded · reconciled: true
observed.inputs.MaximumSpread: 15.0
The parts that are hard to build and easy to get wrong
Validation you do not have to trust the model for
Catalog of EAs with whitelisted inputs, ranges, enum values and presets. The server rejects anything else with a message the model can relay. Your LLM is never the security boundary.
Restart-safety knowledge
Every parameter change restarts the EA. We verify per EA how it recovers its basket and refuse unsafe changes on open positions. Maintained by us as vendors ship new builds.
Reconciliation
Requested, resolved and observed state per command, with timestamps and the inputs read back from the running EA. A mismatch fails the command instead of reporting success.
VPS agent and manager
Outbound-only agent in your VPS image, auto-updating, self-healing. The manager runs as an MQL5 service — nothing a customer can close, no start-up config on the terminal.
Live state and history
Accounts, running EAs with effective settings, positions and orders attributed by magic number, deal history, equity curve and P&L breakdowns — ready for your reporting.
Multi-tenant by design
Tenant → client → VPS → account. Client keys see one customer only. No shared database with us: API only, every call carries the client id.
Backtests as an API call
A pool of dedicated Strategy Tester servers runs backtests for your customers' EAs and settings on request — queued, parsed, returned as structured results your assistant can compare and act on. Your trading VPS machines never run a tester. Specified; scheduled after the pilot.
Three ways in, one contract
Use whichever fits your stack. All three expose the same operations and the same validation.
- REST — OpenAPI 3.1, async commands (202 + command id, long-poll the result)
- Tool definitions — ready-made schemas for Claude and other function-calling models; drop them into your assistant
- MCP server — for MCP-capable clients and internal tooling
| Operation | What it does |
|---|---|
get_client_state | Everything the customer owns, live |
list_eas · get_ea | Catalog with presets, ranges, exact input names |
start_ea · stop_ea | Add or remove a robot, soft or hard stop |
set_risk · set_param · restart_ea | Change settings, dry run first |
get_command | Outcome with diff, observed state, timeline |
get_account_history · get_pnl · get_equity_curve | Deals, P&L breakdowns, equity and balance curves |
From first call to customers in production
Sandbox
A tenant with a simulated VPS and demo accounts. Same validation, dry run and reconcile path as production. Keys within an hour of your request.
Your EA catalog
We onboard each EA from its closed .ex5: inputs are discovered automatically, presets and safe ranges take about ten minutes of review per EA, restart safety is tested on a demo account.
Pilot VPS
Install the agent on one of your VPS images or a test machine with a single script. Discovery finds the terminals; your assistant starts driving them the same day.
Production
Per-customer keys issued from your backend, per-VPS enrolment tokens, monitoring and signed updates from us. Non-exclusive licence, per-VPS pricing.
Request the sandbox
Tell us a little about your platform and which EAs your customers run. You get sandbox keys, the spec and the tool schemas, and a thirty-minute walkthrough if you want one.
Prefer e-mail? info@mtcodec.com