Connect an AI Assistant via MCP

Before you start

1. An AlgoVesta account. The 7-day trial is enough — you do not need a paid plan to test this.
2. An MCP client: Claude on web, desktop or mobile, Claude Code, Cursor, ChatGPT on a paid plan, Gemini CLI, or anything else that speaks Model Context Protocol.
3. A key. You do not need a connected exchange to begin: a paper key runs the full toolset against a $5,000 virtual balance, so you can set everything up and rehearse before any real account is involved.

1. Create your MCP link

1. Open the AlgoVesta panel and go to the MCP Connection tab
2. Click Create key, give it a label that says where it will live (for example claude-desktop or cursor-laptop)
3. Choose a scope: read for portfolio questions only, paper for the full toolset against the $5,000 virtual balance, or live for real orders
4. Copy the address that appears. It looks like this:

https://api.algovesta.com/u/avmcp_<your-key>/mcp

2. Add it to Claude

1. In Claude, open Settings → Connectors
2. Click Add custom connector
3. Give it a name (AlgoVesta) and paste your MCP link into the URL field
4. Save, then start a new conversation — the tools appear in the conversation's tool list

This connector syncs across Claude on web, desktop and mobile, so adding it once covers your phone too. If you would rather not paste a secret address, AlgoVesta also serves a standard OAuth 2.1 endpoint at https://api.algovesta.com/mcp with PKCE (S256) and Dynamic Client Registration — point the connector at that instead and you will be sent through a normal browser consent screen.

3. Add it to Claude Code

One command registers the server over streamable HTTP:

claude mcp add --transport http algovesta https://api.algovesta.com/u/avmcp_<your-key>/mcp

Run claude mcp list afterwards to confirm the server is registered and reachable.

4. Add it to Cursor or ChatGPT

Cursor reads a JSON config file. Add one entry to mcp.json:

{"mcpServers":{"algovesta":{"url":"https://api.algovesta.com/u/avmcp_<your-key>/mcp"}}}

Gemini CLI takes the same entry, but names the field httpUrl instead of url. Everything else is identical.

ChatGPT supports custom MCP connectors on paid plans, with developer mode enabled in settings. Add the connector there and paste the same link.

5. Verify the connection

1. Start a new conversation in your client
2. Ask: How is my portfolio?
3. The assistant should call the get_portfolio_context tool and answer with your actual balances and open positions — on a paper key, with the $5,000 virtual balance

If it answers in generalities without calling a tool, the server is not connected: the assistant is talking about trading rather than about your account. Re-check the link and restart the client.

What the assistant can do once connected

Twenty tools are exposed. Fourteen of them read or rehearse and change nothing: get_portfolio_context, get_market_price, get_trade_history, compare_venues, simulate_order, list_open_orders, list_strategies, verify_receipt, replay_channel, compile_policy, backtest_my_signals, simulate_policy, import_tradingview_backtest and get_job_status. Six of them write: place_order, close_position, modify_position, cancel_order, create_strategy and update_strategy.

You do not call these by name. Ask in plain language — "what am I holding?", "what would 0.1 BTC cost me right now?", "close half my ETH position" — and the assistant picks the tools and chains them. A stop-loss is required on every position it opens, and no tool exists to remove one afterwards. Full schemas for each tool are in the developer documentation.

401 unauthorized

The key was revoked, deleted, or the address is mistyped. This is the most common failure when the link was pasted by hand or copied with a trailing space. Check the key still exists in the MCP Connection tab; if it does, delete it and create a fresh one rather than trying to repair the string.

insufficient_scope

The tool the assistant tried to call needs a higher scope than the key carries. A read key cannot call place_order, and a paper key cannot reach a real account. Scopes are ranked, so a tool requiring paper refuses a read key. Either accept the limit or issue a new key with the scope you actually need — moving to live requires a second factor.

user_frozen

The kill switch is active on your account, which stops every key at once. This is deliberate: it is the emergency brake. Unfreeze it in the panel when you want trading to resume. If you only meant to cut off one client, revoke that single key instead — the others keep working.

rate_limited

You have exceeded 60 requests per minute on that key. Tighter caps apply to the expensive tools: place_order is held to 10 per minute and replay_channel to 5 per hour. Wait and retry. Seeing this repeatedly usually means an agent is looping rather than that you are trading quickly — worth checking what the assistant is actually doing before you raise the request rate.