Agent setup
Use the configured Backplane origin for ChatGPT, Claude Code, and Codex without embedding real credentials.
MCP clients connect to https://backplane.gsmlg.net/mcp; model clients use https://backplane.gsmlg.net/v1. Resource-bound OAuth access tokens are audience-specific, so use separate $MCP_ACCESS_TOKEN and $LLM_ACCESS_TOKEN values. PAT and legacy credentials follow the configured compatibility policy.
Setup
- For ChatGPT, register the exact callback it supplies, select the predefined OAuth client, and configure the MCP endpoint.
- For Claude Code, configure the Anthropic-compatible /v1 base URL and add Backplane as an HTTP MCP server.
- For Codex, define a Backplane model provider and MCP server, with token values supplied through environment variables.
Examples
ChatGPT MCP configuration
MCP server URL: https://backplane.gsmlg.net/mcp
OAuth client ID: $CLIENT_ID
OAuth client secret: $CLIENT_SECRET
Claude Code configuration
export ANTHROPIC_BASE_URL="https://backplane.gsmlg.net"
export ANTHROPIC_AUTH_TOKEN="$LLM_ACCESS_TOKEN"
claude mcp add --transport http --header "Authorization: Bearer $MCP_ACCESS_TOKEN" backplane "https://backplane.gsmlg.net/mcp"
Codex configuration
# Shell
export BACKPLANE_LLM_ACCESS_TOKEN="$LLM_ACCESS_TOKEN"
export BACKPLANE_MCP_ACCESS_TOKEN="$MCP_ACCESS_TOKEN"
# ~/.codex/config.toml
# Replace this placeholder with a model alias exposed by Backplane.
model = "your-backplane-model-alias"
model_provider = "backplane"
[model_providers.backplane]
name = "Backplane"
base_url = "https://backplane.gsmlg.net/v1"
env_key = "BACKPLANE_LLM_ACCESS_TOKEN"
[mcp_servers.backplane]
url = "https://backplane.gsmlg.net/mcp"
bearer_token_env_var = "BACKPLANE_MCP_ACCESS_TOKEN"
Reference
ChatGPT
Claude Code
Codex
MCP /mcp
LLM API /v1