mcp_dynamic_tool_filtering

⚡ Dynamic Tool Filtering in OpenAI Agents SDK ❓ Dynamic Tool Filtering Kya Hai? — Dynamic tool filtering ka matlab hai ke aap apne agent ko dene waali tools ki list ko run-time par control kar sakte hain.

Matlab: Aap ek function likhte hain jo decide karta hai ke is waqt agent ke liye konsi tools available honi chahiye aur konsi nahi.

Ye ek “smart menu” ki tarah hai:

Subah → breakfast menu

Raat → dinner menu

Bilkul isi tarah, agent ka tool menu bhi context ke hisaab se change hota hai.

🔑 Ye Zaroori Kyun Hai?

🧠 Context-aware control — Agent ko uske kaam ke mutabiq tools milte hain.

Example: Agar agent sirf mood-related kaam kar raha hai → sirf mood tools show hongi.

🔒 Behtar Security — Aap decide karte ho kon user/agent kis tool ko access kar sakta hai.

Sensitive tools ko hide kiya jaa sakta hai.

⚡ Flexibility

Static filtering mein → ek baar jo tools allow kar diye, wahi fix rahte hain.

Dynamic filtering mein → tools ki list conditions ke hisaab se change hoti hai.

⚙️ Ye Kaam Kaise Karta Hai? 1️⃣ Ek Filter Function Banayein — Filter function do cheezen leta hai:

ToolFilterContext → agent aur server ki info deta hai.

tool → har tool ke details.

Function return karega:

True → Tool available hogi

False → Tool hide kar di jayegi

2️⃣ Example Logics 🔹 Simple Logic return tool.name.startswith(“mood”)

👉 Sirf woh tools show hongi jo “mood” se shuru hoti hain.

🔹 Context-aware Logic return context.agent.name == “MyMCPConnectedAssistant” and tool.name == “mood_from_shared_server”

👉 Agar agent ka naam “MyMCPConnectedAssistant” hai, tab sirf “mood_from_shared_server” tool show hogi.

🔹 Async Logic

Aap async function bhi bana sakte ho → kisi API ya database check ke basis par decide karne ke liye.

3️⃣ MCP Server Client Mein Use Karein from agents.mcp import MCPServerStreamableHttp, MCPServerStreamableHttpParams from agents.mcp.types import ToolFilterContext

Apna filter function

def custom_filter(context: ToolFilterContext, tool) -> bool: # Example logic: sirf mood wali tools show karo return tool.name.startswith(“mood”)

mcp_params = MCPServerStreamableHttpParams(url=”http://localhost:8000/mcp/”)

async with MCPServerStreamableHttp( params=mcp_params, tool_filter=custom_filter, name=”MyDynamicMCPServer” ) as mcp_server_client: # Ab agent run karein, woh sirf mood wali tools use kar payega pass — 🎯 Final Result — Dynamic tool filtering agent ko:

Smarter banata hai 🧠

Secure banata hai 🔒

Fine-grained control deta hai ⚙️

👉 Isse aap apne agent ki abilities ko specific task aur situation ke hisaab se customize kar sakte ho.