> ## Documentation Index > Fetch the complete documentation index at: https://docs.magichour.ai/llms.txt > Use this file to discover all available pages before exploring further. # DSPy > Check an existing Magic Hour image project with DSPy's native MCP tools, without a language-model call or another generation. Connect DSPy's native MCP tools to Magic Hour, check API authorization, and recover metadata for an image project you already created. This example makes no model, creation, or media-download request. ## Install and configure Requires Python 3.12. Download [recover-image.py](https://raw.githubusercontent.com/magichourhq/docs/main/examples/dspy/recover-image.py) and [requirements.txt](https://raw.githubusercontent.com/magichourhq/docs/main/examples/dspy/requirements.txt) into the same directory, then install in a virtual environment: ```bash theme={null} python3.12 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt ``` The example pins `dspy[mcp]` to 3.4.1, `mcp` to 2.3.0, and `httpx2` to 2.16.0. Provide a [Magic Hour API key](https://magichour.ai/developer?tab=api-keys) through your server-side secret manager as `MAGIC_HOUR_API_KEY`. Set `MAGIC_HOUR_IMAGE_PROJECT_ID` to the saved ID returned by your original image-creation request. Keep keys, account details, and temporary download URLs private. ## Check the saved project ```bash theme={null} python recover-image.py ``` DSPy's `Tool.from_mcp_tool` converts discovered tools into native DSPy tools. Calling `.acall()` directly invokes the selected tool without an LM or agent choosing what to call. Discovery alone does not validate your key; the example checks account field types before reporting authorization. ```python theme={null} import asyncio import json import os async def main(): phase = "configure" try: import dspy import httpx2 from mcp import ClientSession from mcp.client.streamable_http import streamable_http_client dspy.configure(callbacks=[]) dspy.configure_cache(enable_disk_cache=False, enable_memory_cache=False) project_id = os.environ["MAGIC_HOUR_IMAGE_PROJECT_ID"] phase = "connect" async with httpx2.AsyncClient( headers={"Authorization": f"Bearer {os.environ['MAGIC_HOUR_API_KEY']}"}, follow_redirects=False, timeout=60, ) as http_client: async with streamable_http_client( "https://mcp.magichour.ai/", http_client=http_client ) as (read, write): async with ClientSession(read, write) as session: await session.initialize() phase = "convert_tools" available = await session.list_tools() tools = { tool.name: dspy.Tool.from_mcp_tool( session, tool, result_mode="structured" ) for tool in available.tools } print({"tool_count": len(tools)}) phase = "account" account = await tools["account_retrieve"].acall() if isinstance(account, str): account = json.loads(account) if not isinstance(account, dict): raise ValueError("Unexpected account result type") if not isinstance(account["id"], str) or type(account["credits"]) not in (int, float): raise ValueError("Unexpected account field types") print({"account_authenticated": True}) phase = "project" project = await tools["wait_for_image_project"].acall( id=project_id, timeout_seconds=30, include_inline_downloads=False, ) if isinstance(project, str): project = json.loads(project) if not isinstance(project, dict): raise ValueError("Unexpected project result type") if project["id"] != project_id or not isinstance(project["status"], str): raise ValueError("Unexpected project fields") print({"id": project["id"], "status": project["status"]}) phase = "cleanup" return 0 except Exception: print({"error": "DSPy MCP example failed", "phase": phase}) return 1 if __name__ == "__main__": raise SystemExit(asyncio.run(main())) ``` The example prints the discovered tool count, an authorization marker, and the requested project's ID and status. It does not print the account balance, credentials, raw responses, or signed media URLs. Each HTTP client and MCP session closes through its native async context. DSPy caches are disabled, callbacks are empty, and no model or tracer is configured. ## Results and recovery `result_mode="structured"` uses MCP structured content when available. DSPy's fallback may return a JSON string; the example accepts a dictionary or that JSON-string form. It checks that the returned project ID matches the saved ID and that status is a string. The wait tool may poll the same saved project for up to 30 seconds. A timeout does not mean generation failed or stopped. Retain the saved ID and run recovery later. On failure, the example reports only a static phase label and exits with status 1. It never creates another project or automatically retries a creation request. This example recovers metadata only. Inline downloads are disabled. DSPy 3.4.1's MCP conversion can prefer text over non-text content when both are present; this example does not preserve inline images or save a media file. Model reasoning and image creation require separate application logic and are outside its scope. ## Scope and verification Native validation with Python 3.12, DSPy 3.4.1, and MCP 2.3.0 converted 48 tools, verified account field types, and recovered the saved image project with status `complete`. The exact runnable example exited successfully with no stderr; its client contexts closed and the process ended. No model, creation, or media-download call was made. Discovery does not mean all 48 tools were executed. See [Magic Hour MCP](https://docs.magichour.ai/integration/model-context-protocol.md), our [privacy policy](https://magichour.ai/privacy-policy), and [DSPy's official MCP tools documentation](https://dspy.ai/3.4.1/diving-deeper/tools/). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.