Using AI Coding Agents with Volary
Volary adds persistent memory to your coding tools. Your agent learns your codebase, your conventions, and your preferences – then applies that knowledge automatically in every session.
Supported tools
Claude Code
Anthropic's CLI coding agent. Point it at Volary with two environment variables.
OpenAI Codex
OpenAI's coding agent. Configure Volary as a model provider in your Codex config.
Cursor
AI-powered code editor. Connect to Volary via MCP server configuration and Cursor rules.
Getting started
Before configuring your coding tool, set up Volary:
- 1Sign up or log in to the Volary dashboard.
- 2Create an organisation and an agent. The setup wizard will guide you through choosing your tool and configuring an upstream endpoint.
- 3Copy the configuration shown in the setup wizard, or follow the tool-specific instructions below.
Claude Code
Paste the exports below into your shell, then run claude. Replace the placeholder values with your organisation ID, agent ID, and API token from the dashboard.
export ANTHROPIC_BASE_URL="https://api.volary.ai/v0/orgs/{ORG_ID}/agents/{AGENT_ID}"
export ANTHROPIC_CUSTOM_HEADERS="X-Volary-Api-Key: {YOUR_TOKEN}"Your Anthropic API key stays local – Volary uses the upstream endpoint you configured in the dashboard.
OpenAI Codex
Add the following to your Codex config file (usually ~/.codex/config.toml). Replace the placeholder values with your details from the dashboard.
[model_providers.volary]
name = "volary memory server"
base_url = "https://api.volary.ai/v0/orgs/{ORG_ID}/agents/{AGENT_ID}/v1"
http_headers = { "X-Volary-Api-Key" = "{YOUR_TOKEN}" }
env_key = "OPENAI_API_KEY"
wire_api = "responses"
[profiles.volary]
model_provider = "volary"
profile = "volary"Cursor
Open ~/.cursor/mcp.json (or .cursor/mcp.json for project-specific config) and add the following. Replace the placeholder values with your details from the dashboard.
{
"mcpServers": {
"volary": {
"url": "https://api.volary.ai/v0/orgs/{ORG_ID}/agents/{AGENT_ID}/v0/mcp",
"headers": {
"Authorization": "Bearer {YOUR_TOKEN}"
}
}
}
}Then add two cursor rules (user or project rules):
volary transcribe
After finishing any user request, you MUST call the transcribe tool on the volary MCP server exactly once.
transcribe.messages should contain the transcript of messages sent back and forth between the user and the model backend, including all the tool call responses etc.
Never include credentials, tokens or private keys in the call to transcribe.volary recall
When starting any user request, you MUST call the recall tool on the volary MCP server exactly once.
This may return you relevant memories about past experiences when something similar has been done, which may inform you better about how to proceed this time.Ready to get started? Sign up and create your first agent, or check the API guide for programmatic integration.