diff --git a/README.md b/README.md
index 728ffab4..6b7618e8 100644
--- a/README.md
+++ b/README.md
@@ -59,6 +59,20 @@ pip install -e ".[dev]"
 
 For a more detailed first-time setup flow, including MCP server verification and troubleshooting, see [Beginner Setup](#-beginner-setup-firsttime-users).
 
+## 🔌 Integrating with External Agents
+
+If you are building an external agent framework (e.g. LangChain, AutoGen, or a custom ReAct loop) and want to import `sktime-mcp` tools directly into your Python codebase without running the MCP server, you must ensure the package is properly installed.
+
+**Local Development Warning:** Do not simply clone the repository and try to import `sktime_mcp` from a sibling directory without installing it. The package relies on namespace resolution. 
+
+Always install the package in editable mode from the repository root:
+```bash
+pip install -e .
+```
+*(Alternatively, explicitly set `PYTHONPATH=.` when running your agent script).*
+
+For a complete runnable example of how a custom Python agent should load data and evaluate an estimator using the raw API, see [`examples/external_agent_integration.py`](examples/external_agent_integration.py).
+
 ## 🧭 Beginner Setup (First‑Time Users)
 
 If you are new to sktime‑mcp or to MCP‑based workflows, this section provides a minimal starting point to help you verify that your setup is working correctly.
diff --git a/examples/external_agent_integration.py b/examples/external_agent_integration.py
new file mode 100644
index 00000000..1f0e6e81
--- /dev/null
+++ b/examples/external_agent_integration.py
@@ -0,0 +1,110 @@
+#!/usr/bin/env python
+"""
+Example: External Agent Integration Workflow
+
+This script demonstrates how an external LLM agent (or its backend)
+should programmatically interact with the `sktime_mcp` server tools
+when using the raw module API.
+
+It covers the complete workflow:
+1. Loading custom data
+2. Reading returned metadata
+3. Instantiating a forecaster
+4. Evaluating the forecaster on the loaded data
+
+Run this example:
+    python examples/external_agent_integration.py
+"""
+
+import sys
+import json
+import tempfile
+from pathlib import Path
+import pandas as pd
+
+sys.path.insert(0, "src")
+
+from sktime_mcp.tools.data_tools import load_data_source_tool
+from sktime_mcp.tools.instantiate import instantiate_estimator_tool
+from sktime_mcp.tools.evaluate import evaluate_estimator_tool
+
+def print_step(title: str):
+    print("\n" + "=" * 60)
+    print(f"  {title}")
+    print("=" * 60)
+
+def main():
+    print("\n🤖 External Agent Integration Workflow Demo")
+    
+    # Create some dummy external data
+    db_path = Path(tempfile.gettempdir()) / "agent_dummy_data.csv"
+    sample_data = pd.DataFrame({
+        "date": pd.date_range(start="2020-01-01", periods=100, freq="D"),
+        "sales": [100 + i + (i % 7) * 5 for i in range(100)]
+    })
+    sample_data.to_csv(db_path, index=False)
+    
+    # ---------------------------------------------------------
+    # STEP 1: Load Custom Data
+    # ---------------------------------------------------------
+    print_step("STEP 1: Load Custom Data")
+    print(f"Loading data from {db_path}...")
+    
+    data_config = {
+        "type": "file",
+        "path": str(db_path),
+        "time_column": "date",
+        "target_column": "sales"
+    }
+    
+    load_res = load_data_source_tool(data_config)
+    if not load_res["success"]:
+        print("Failed to load data:", load_res)
+        return
+        
+    data_handle = load_res["data_handle"]
+    print("✅ Data loaded successfully.")
+    print(f"Handle assigned: {data_handle}")
+    print("Metadata:", json.dumps(load_res["metadata"], indent=2))
+    
+    # ---------------------------------------------------------
+    # STEP 2: Instantiate Forecaster
+    # ---------------------------------------------------------
+    print_step("STEP 2: Instantiate Forecaster")
+    
+    print("Instantiating NaiveForecaster...")
+    inst_res = instantiate_estimator_tool("NaiveForecaster", params={"strategy": "last"})
+    
+    if not inst_res["success"]:
+        print("Failed to instantiate:", inst_res)
+        return
+        
+    estimator_handle = inst_res["handle"]
+    print("✅ Forecaster instantiated successfully.")
+    print(f"Handle assigned: {estimator_handle}")
+    
+    # ---------------------------------------------------------
+    # STEP 3: Evaluate Estimator
+    # ---------------------------------------------------------
+    print_step("STEP 3: Evaluate Estimator on Custom Data")
+    
+    print("Running 3-fold cross-validation...")
+    eval_res = evaluate_estimator_tool(
+        estimator_handle=estimator_handle,
+        data_handle=data_handle,
+        cv_folds=3
+    )
+    
+    if not eval_res["success"]:
+        print("Failed to evaluate:", eval_res)
+        return
+        
+    print("✅ Evaluation completed successfully.")
+    print("\nMetrics:")
+    for metric, value in eval_res["metrics"].items():
+        print(f"  - {metric}: {value:.4f}")
+        
+    print("\nWorkflow completed successfully!")
+
+if __name__ == "__main__":
+    main()