> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heliyos.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Configuration

> Environment variables and the tool timeout the MCP server needs.

## Environment variables

| Variable | Required | Default | What it does |
| - | - | - | - |
| `HELIYOS_API_KEY` | Yes | None | Your Heliyos API key. It must start with `heliyos_`. |
| `HELIYOS_BASE_URL` | No | `https://api-prod.heliyos.ai` | The Heliyos API to call. It must use HTTPS. |

## Tool timeout

`heliyos_ask` waits up to 90 seconds for Heliyos to answer, so give tool calls at least 120 seconds.

| Client | Setting |
| - | - |
| Codex | `tool_timeout_sec = 120` in `[mcp_servers.heliyos]` (the default is 60). A plugin installed from a marketplace keeps the 60 second default; use the `config.toml` setup for 120. |
| Claude Code | The default is long enough. If you set `MCP_TOOL_TIMEOUT`, keep it at `120000` or more. |
| Claude Desktop, Cursor, VS Code, Antigravity | No documented setting. |

When a client cuts a call short, the run keeps going in Heliyos. Find it with `heliyos_runs_list` and follow it with `heliyos_runs_get` and `wait: true`, which waits up to 30 seconds per call.

A rate-limited request (HTTP 429) is retried once after the `Retry-After` delay, capped at 30 seconds, so a `heliyos_ask` that hits the limit can take about 125 seconds. If you see timeouts near the limit, raise the timeout to 150 seconds.

## Response size

Reads take `response_format`: `concise` (the default) returns a short summary, `detailed` returns everything. Lists take `limit` (1 to 100) and `cursor`. Large answers are cut with a note telling the agent how to ask for less.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.