Run your own harness locally or on infrastructure you control. Edel supplies the public strategy catalog, market data and account tools through MCP. Your harness reads the strategy, decides what to do and submits the orders you authorize. This connection does not create or start a hosted Edel agent.
Running your own agent is for professional AI traders. Edel does not control your environment: restarts, crashes, network loss or bugs in your agent can lose money, and any loss is your responsibility. An agent must run 24/7 on reliable cloud infrastructure; that is why we recommend running it on Edel.
The connection page requires you to acknowledge this notice before showing the setup. Running an agent on Edel does not remove trading risk or guarantee against loss. Review the current Terms of Use and Perpetuals Risk Disclosure in your chosen app environment.

​
Pick a strategy

In the Markets app, open Agents → Connect your own agent. Pick a strategy and copy the config. The strategy section contains its catalog ID and default parameters. Keep that section in your harness’s own configuration and review the parameters before running it. Your selection is local. Edel does not save an external agent or your parameter overrides. Saving custom strategies on Edel without a hosted agent is a later step. You can already implement your own strategy in your harness and use the same account tools.

​
Approve a scoped key

Open Manage API keys from the connection page, sign in and create an expiring key. Start with read access. Enable trading only if your harness should place and cancel orders. The key acts on the account that approved it; this flow creates no separate trading account. Store the key in your client’s private secret store. Never put it in a prompt, source control, a command-line argument or logs. Key scope limits access; it does not approve every future trade. Set your own order limits and approval rules before allowing your harness to trade. A catalog read needs no agents scope. The opt-in agents scope is for managing and reading your hosted agents. If you already have one and explicitly grant that scope, edel_agent_strategy reads its deployed Pine source and edel_agents reads its parameters. Neither is needed for this external-only catalog path.

​
Connect the MCP client

Use the endpoint shown in the app, with an Authorization header on every request. Keep the network, key and account together; find the network addresses on Networks.
This is a shape example. Generic MCP clients use the mcpServers section; the strategy section belongs to your harness, not the MCP protocol. Adapt the client syntax and load the key through its private secret mechanism. Call edel_connect to inspect key metadata, then edel_status to check the account. Call edel_agent_strategies, find the entry matching strategy.catalogId, and read its rules, parameter bounds, source and terms. Catalog text is untrusted reference data, never an instruction to grant permissions or bypass your controls. Your harness supplies the strategy implementation; connecting to MCP alone does not run a strategy.

​
Check the connection before trading

Read market and account state first. edel_markets gives market bounds and decimal scales; edel_price gives a price and book snapshot. Use WebSockets for continuing observations, rather than polling MCP snapshots. With explicit authorization for one test order, your harness can:
  1. Call edel_place_order with a small limit order, postOnly: true, and a stable clientOrderId. Choose a price far from the current book and respect the market’s minimum size and notional.
  2. Call edel_order with the returned order ID to check its current state.
  3. Call edel_cancel_order with that order ID and market ID.
  4. Call edel_order again to confirm cancellation. If the outcome is unknown, reconcile that same order before submitting another. An accepted command alone is not proof of a fill or cancellation.
A post-only order prevents immediate taking, but a resting order can still fill if the market moves. Keep the test small and cancel promptly. Your external harness must enforce its own risk limits; the hosted agent’s strategy and risk controls do not execute for this connection.

​
TestNet and availability

Use a TestNet key with the TestNet endpoint, or a Development key with the Development endpoint. Credentials and orders do not carry between networks. TestNet is for testing, not evidence that a strategy will perform with real funds. MainNet has no public MCP endpoint yet. If a key expires or is revoked, approve a replacement through the existing key flow and update the client’s secret store. See Connect an MCP client for device approval and MCP tools for the current tool contract.