{
  "id": "claude-code",
  "name": "Claude Code",
  "category": "coding-agent",
  "recipe": "claude-env",
  "protocol": "messages",
  "source": "https://code.claude.com/docs/en/third-party-integrations",
  "referenceDate": "2026-09-29",
  "verification": {
    "setup": "prepared",
    "session": "not_run"
  },
  "notes": "Use a dedicated Jesse launch environment. Recorded Jesse runs exist; this release has not rerun the installed CLI.",
  "contractVersion": "1.0.0",
  "endpoint": "https://jesse.my/api/v1/messages",
  "apiRoot": "https://jesse.my/api",
  "model": "jesse",
  "files": [
    {
      "name": "launch-jesse.sh",
      "format": "shell",
      "content": "#!/bin/sh\nset -eu\n: \"${JESSE_API_KEY:?Set your Jesse key in the launch environment}\"\nexport ANTHROPIC_BASE_URL='https://jesse.my/api'\nANTHROPIC_AUTH_TOKEN=\"$JESSE_API_KEY\"\nexport ANTHROPIC_AUTH_TOKEN\nunset ANTHROPIC_API_KEY\nexport ANTHROPIC_MODEL='jesse'\nexport ANTHROPIC_SMALL_FAST_MODEL='jesse'\nexec claude \"$@\"\n"
    },
    {
      "name": "README.md",
      "format": "markdown",
      "content": "# Claude Code with Jesse\n\nUse a dedicated Jesse launch environment. Recorded Jesse runs exist; this release has not rerun the installed CLI.\n\nAPI root: https://jesse.my/api\nOpenAI-compatible base URL: https://jesse.my/api/v1\nEndpoint: https://jesse.my/api/v1/messages\nModel: jesse\nKey environment: JESSE_API_KEY\n\nKeep the key in the app's credential store or launch environment. These files are merge snippets; never overwrite existing providers or settings. GUI apps may not inherit shell environment variables.\n\nClient budgets: context 10000, output 4096. These are conservative defaults, not server capacity claims. Keep general vision, embeddings, audio, provider-hosted web tools and automatic retry of billable turns disabled.\n\nFull Jesse API access: shared/client_sdk.mjs supports 31 operations. The MCP bridge exposes the same operation list where the host supports MCP. Preserve the full Jesse receipt. Tool calls remain proposals and are not executed by the bridge.\n\nVerification: setup prepared; current full application session not_run. Run a chat, a streamed reply, a refused credential, a failing tool result and a file task with its actual tests before qualifying a coding-agent release.\n\nUpstream: https://code.claude.com/docs/en/third-party-integrations\n"
    }
  ]
}