Zia AI — Customer Support Connector

Ye aapka apna MCP connector hai. Iska kaam: customer ke FAQs ka jawab dena, order ka status batana, aur har customer ko yaad rakhna (agli chat mein bhi).

Neeche diye steps apne computer par karne hain (jahan internet ho). Har step ke baad check kar lein ke wo kaam ho gaya, phir agle par jayein.

Step 0 — Cheezein jo chahiye

• Python 3.10 ya us se naya (terminal mein check: python3 --version)

• Internet connection

• (Optional, baad mein) Claude Code ya koi bhi terminal/coding agent

Agar Python install nahi hai, python.org se install kar lein.

Related MCP server: Enneagora E-commerce MCP Server

Step 1 — Files nikalein (unzip karein)

Jo zip file di gayi hai usay kisi folder mein unzip karein, phir terminal us folder ke andar khol kar cd karein. Misal:

cd Downloads
unzip zia-ai-connector.zip
cd zia-ai-connector

Step 2 — Packages install karein

pip install -r requirements.txt

Agar pip na chale, pip3 try karein. Agar permission error aaye:

pip install -r requirements.txt --break-system-packages

Step 3 — Settings file banayein

cp .env.example .env

Is .env file ko kisi bhi text editor mein khol kar dekh lein — abhi kuch badalne ki zaroorat nahi, defaults theek hain demo ke liye.

Step 4 — Apne computer par chala kar dekhein

python run.py

Agar sahi chala to terminal mein kuch aisa dikhega:

Zia AI connector shuru ho raha hai...
AUTH_DISABLED = 1

aur server chalta reh jayega (band mat karein, isi terminal ko khula rehne dein). Ye is baat ka saboot hai ke aapka connector zinda hai. Band karne ke liye Ctrl+C dabayein.

Agar koi error aaye: error message copy kar ke apne coding agent (Claude Code) ko dikhayein aur kahein "is error ko thik karo" — ye normal hai, kabhi kabhi packages ka version thora alag hota hai.

Step 5 — Internet par live karein (tunnel)

Abhi tak ye server sirf aapke apne computer ke andar chal raha hai — claude.ai is tak nahi pohanch sakta. Usay internet par lana hai. Sab se asan tareeqa Cloudflare tunnel hai (free, login ki zaroorat nahi):

1. Naya terminal tab kholein (purana wala jahan server chal raha hai, usay chalta rehne dein).

2. Cloudflared install karein:

   • Mac: brew install cloudflared

   • Windows: cloudflared download page se .exe download karein

   • Linux: sudo apt install cloudflared ya releases page se download karein

3. Ye command chalayein (server jis port par chal raha hai, default 8000):

   cloudflared tunnel --url http://localhost:8000

4. Terminal mein ek URL print hoga, jaisa:

   https://random-words-1234.trycloudflare.com

   Ye URL copy kar lein — agle step mein chahiye hoga. Iske aakhir mein /mcp lagana hoga: https://random-words-1234.trycloudflare.com/mcp

Is terminal ko bhi khula/chalta rehne dein.

Step 5.1 — Docker par run karna (optional deploy)

Agar aap local system par Docker install karte hain, to ye connector container mein chala sakte hain:

cd zia-ai-connector
docker build -t zia-ai-connector .
docker run --rm -p 8000:8000 --env-file .env zia-ai-connector

Phir aapka connector browser se http://localhost:8000 par chalta hai. Tunnel chalane ke liye wohi cloudflared tunnel --url http://localhost:8000 use karein.

Agar aap cloud host (Railway/Render/Azure) par deploy karna chahein, to bas ye Docker image push karein ya python run.py ko service ke roop mein chalaen. Ensure .env mein AUTH_DISABLED=1 demo mode ke liye set ho.

Render users: repository root mein render.yaml file bana di hai. Agar aap Render dashboard par hain, repo connect karne ke baad ye file auto detect ho sakti hai.

Render deploy steps:

1. Render.com par account login karo.

2. New → Web Service.

3. GitHub repo select karo (iss project ka repo).

4. Build command waisa hi rehne do: pip install -r requirements.txt.

5. Start command set karo: python run.py.

6. Environment variables ke liye AUTH_DISABLED=1 aur AUTH_ISSUER=zia-ai-mock-auth add karo.

7. Deploy par click karo.

Agar aapki repo private hai, to Render ko permissions do repo access karne ke liye.

Step 6 — claude.ai mein connector add karein

1. claude.ai kholein aur sign in karein.

2. Settings → Connectors → Add custom connector par jayein.

3. Wahi URL paste karein jo Step 5 mein mila tha (/mcp ke sath).

4. Add par click karein.

Bas! Koi Authorize step nahi chahiye is demo mein (kyunke AUTH_DISABLED=1 hai).

Step 7 — Test karein

Naya chat kholein aur Claude se kahein:

"Mera order ORD-1001 ka status pata karo"

Ya:

"Return policy kya hai?"

Claude khud begin_session call karega, phir aapka connector use kar ke jawab dega. Ek naya chat khol kar dobara pucha:

"Maine pehle kya pucha tha?"

— to ye purani baat yaad rakhega, kyunke data customer ki id (sub) ke sath save hota hai, chat ke sath nahi.

Important baatein (zaroor parhein)

• AUTH_DISABLED=1 ka matlab: filhal har koi "demo-customer-1" ki tarah dikhega — ye sirf demo/testing ke liye theek hai. Asal mein alag-alag logon ko alag pehchanwane ke liye real sign-in service (jaise Clerk ya Auth0) chahiye hogi — ye agla qadam hai.

• Tunnel URL har baar badalta hai jab aap cloudflared dobara chalate hain. Naya URL milne par claude.ai mein connector dobara add karna paray ga (purana hata kar).

• Jab kaam khatam ho, dono terminal (python run.py aur cloudflared) Ctrl+C se band kar dein.

• Ye database (zia_ai.db) aapke computer par hi save hoti hai. Agar delete kar dein to data reset ho jayega.

Aage kya?

• Apna khud ka FAQ/orders data chahiye? seed/faqs.json aur seed/orders.json files edit kar lein.

• Real, alag-alag customers ke liye real sign-in chahiye? Ye agla course hai jo guide mein "AI Identity" kehlata hai.

• Hamesha ke liye chalta rahe (computer band hone par bhi)? Isay kisi cloud host (Railway, Render, ya Azure) par "deploy" karna hoga — abhi ke liye laptop + tunnel kaafi hai seekhne ke liye.