Claude Code

Use mercury-cc to launch Claude Code with Mercury 3. The launcher runs an isolated local bridge, so your regular claude command and Claude Code settings stay unchanged.

This integration uses LiteLLM as a third-party bridge between Claude Code’s Anthropic Messages API and Inception’s OpenAI-compatible API. Anthropic does not support routing Claude Code to non-Claude models. Review LiteLLM before using it with sensitive code or credentials.

Mercury 3 supports parallel tool calls, allowing Claude Code to request multiple independent tools in the same turn.

Prerequisites

Claude Code
macOS or Linux
Python 3.11 or later
An Inception API key

Setup

Install the local bridge

Set PYTHON_BIN to a Python 3.11+ executable, then install the tested LiteLLM version in an isolated environment:

PYTHON_BIN=python3
"$PYTHON_BIN" --version

mkdir -p "$HOME/.mercury-cc" "$HOME/.local/bin"
chmod 700 "$HOME/.mercury-cc"
"$PYTHON_BIN" -m venv "$HOME/.mercury-cc/.venv"
"$HOME/.mercury-cc/.venv/bin/pip" install 'litellm[proxy]==1.99.0'

Store your Inception API key

Enter your API key when prompted. The key stays in a user-only file and is only loaded by the local bridge.

bash -c 'read -rsp "Inception API key: " key; printf "\n"; printf "%s" "$key" > "$HOME/.mercury-cc/api-key"'
chmod 600 "$HOME/.mercury-cc/api-key"

Create the mercury-cc launcher

Create ~/.mercury-cc/mercury-cc:

#!/usr/bin/env bash
set -euo pipefail

mercury_home="$HOME/.mercury-cc"
key_file="${mercury_home}/api-key"
pid_file="${mercury_home}/proxy.pid"
port="${MERCURY_CC_PORT:-4000}"
proxy_url="http://127.0.0.1:${port}"

proxy_ready() {
  curl -fsS "${proxy_url}/v1/models" 2>/dev/null \
    | grep -q 'inception/mercury-3'
}

if [[ "${1:-}" == "--stop" ]]; then
  if [[ ! -f "$pid_file" ]]; then
    echo "Mercury bridge is not running."
    exit 0
  fi

  read -r pid < "$pid_file"
  if kill -0 "$pid" 2>/dev/null; then
    kill "$pid" 2>/dev/null || true
    echo "Mercury bridge stopped."
  else
    echo "Mercury bridge is already stopped."
  fi
  rm -f "$pid_file"
  exit 0
fi

if [[ ! -s "$key_file" ]]; then
  echo "Inception API key not found at ${key_file}" >&2
  exit 1
fi

if ! proxy_ready; then
  INCEPTION_API_KEY="$(<"$key_file")" \
    nohup "${mercury_home}/.venv/bin/litellm" \
      --model inception/mercury-3 \
      --alias mercury \
      --drop_params \
      --host 127.0.0.1 \
      --port "$port" \
      >"${mercury_home}/proxy.log" 2>&1 &
  echo "$!" > "$pid_file"

  for _ in {1..60}; do
    proxy_ready && break
    sleep 0.25
  done
fi

if ! proxy_ready; then
  echo "Mercury bridge failed to start. See ${mercury_home}/proxy.log" >&2
  exit 1
fi

export ANTHROPIC_BASE_URL="$proxy_url"
export ANTHROPIC_AUTH_TOKEN="mercury-local"
unset ANTHROPIC_API_KEY

export ANTHROPIC_MODEL="mercury"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="mercury"
export ANTHROPIC_DEFAULT_SONNET_MODEL="mercury"
export ANTHROPIC_DEFAULT_OPUS_MODEL="mercury"
export CLAUDE_CODE_SUBAGENT_MODEL="mercury"
export CLAUDE_CODE_MAX_CONTEXT_TOKENS=1000000

exec claude "$@"

Make it executable and link it into ~/.local/bin:

chmod 700 "$HOME/.mercury-cc/mercury-cc"
ln -sf "$HOME/.mercury-cc/mercury-cc" "$HOME/.local/bin/mercury-cc"

If ~/.local/bin is not already on your PATH, link the launcher into another user-writable directory that is. This setup does not modify your shell configuration.

Launch Claude Code with Mercury

Start an interactive session:

mercury-cc

Run /status inside Claude Code to confirm that the API base URL is http://127.0.0.1:4000 and the model is mercury.

Verify the integration

Run an end-to-end prompt:

mercury-cc -p "Reply with exactly: Mercury 3 is connected."

You can also call the bridge’s Anthropic-compatible endpoint directly:

curl http://127.0.0.1:4000/v1/messages \
  -H 'Authorization: Bearer mercury-local' \
  -H 'Anthropic-Version: 2023-06-01' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "mercury",
    "max_tokens": 2048,
    "messages": [{"role": "user", "content": "Say hello from Mercury 3."}]
  }'

This request uses max_tokens because it targets the Anthropic Messages interface exposed by the local bridge. Direct requests to Inception’s Chat Completions API use max_completion_tokens.

Stop the bridge

mercury-cc --stop

Troubleshooting

Bridge startup fails: Run tail -n 50 ~/.mercury-cc/proxy.log and confirm that Python is version 3.11 or later.
mercury-cc is not found: Run ~/.mercury-cc/mercury-cc directly or link it into a directory already on your PATH.
Claude Code logs unrecognized_model: Claude Code may show this warning for the custom mercury alias. Requests still route to Mercury 3.
Regular Claude Code behavior: Launching claude directly continues to use your existing Claude Code configuration.

Claude Code gateway documentation · LiteLLM Inception provider

Getting Started

API reference

Capabilities

Guides

Resources

Support

Prerequisites

Setup

Verify the integration

Stop the bridge

​Prerequisites

​Setup

​Verify the integration

​Stop the bridge

Prerequisites

Setup

Verify the integration

Stop the bridge