# Phase 1 — Local MCP server & first tool

## Goal
Run the MCP server locally, expose a health check, and implement basic tools/resources/prompts.

## What you get
- `server.py` with tools: `add`, `echo`
- Resource: `prynai://status`
- Prompt: `quick_summary` (simple template)
- Local smoke tests under `examples/`

## Prereqs
- Python 3.11+ and `uv`
- (Optional) Docker if you prefer containers

## Setup
```bash
cp .env.example .env
# For Phase 1 local runs you can temporarily leave OAuth off or point to dev values.
uv sync

# Phase 2 — Redis state (counter)

## Goal
Add Redis to persist state and expose a counter as an MCP resource/tool.

## What you get
- Redis connection (via `REDIS_URL` or host/port/password)
- `bump_counter` tool
- `prynai://counter` resource

## Prereqs
- A Redis instance (local or cloud)
- Update `.env` with Redis settings

## Run locally
bash
# Option A: docker compose (includes Redis)
docker compose up --build

# Option B: your own Redis, run server with uv
uv run python examples/use_core_list_tools.py
uv run python examples/use_core_langchain_agent.py

# Phase 3 — OAuth (Entra ID client credentials)

## Goal
Protect `/mcp` with Entra ID (Azure AD) OAuth2 (client credentials flow).

## What you get
- Server validates `Authorization: Bearer <token>`
- Example client code to fetch tokens via `msal`

## Configure Entra
- **App registrations**: one for the server (expose API), one for the client
- Record:
  - `ENTRA_TENANT_ID`
  - `ENTRA_CLIENT_ID`, `ENTRA_CLIENT_SECRET` (client app)
  - `SERVER_APP_ID_URI` (e.g., `api://<server-app-id>`)

Update `.env`:

# Phase 4 — Azure Container Apps deployment

## Goal
Deploy the MCP server to **Azure Container Apps (ACA)**, use **Azure Container Registry (ACR)** for images, and wire **Redis** & OAuth secrets.

## Azure resources (example names)
- RG: `prynai-mcp-rg`
- ACA Env: `prynai-aca-env`
- Container App: `prynai-mcp` → `https://<fqdn>/mcp`
- ACR: `prynaiacr44058.azurecr.io`
- Azure Cache for Redis: e.g., `prynai-redis.redis.cache.windows.net`

## First-time deploy
Use `infra/azure/deploy_aca.ps1` (provisions env/app + initial image). Fill environment in the script or pass parameters as needed.

## Updating (safe rollout)
Use **remote build** + **revision traffic** via:
```powershell
.\infra\azure\deploy_update.ps1 -Tag 0.5.0-phase5
# Script prints a new revision FQDN for smoke tests and the promote command.

### What the update script does:

- Builds a new image in ACR (no local Docker needed).

- Creates a new ACA revision with that image.

- Keeps multiple revisions enabled for safe testing.

- You manually promote traffic to 100% when satisfied.


## Promote to 100%

### The script prints the exact az containerapp ingress traffic set ... command with the new revision name.

## Troubleshooting

- If a revision gets a new sub-FQDN, that’s expected in “multiple revisions” mode. Your stable app URL stays the same once you promote.


---

### `docs/phase5.md`

```markdown
# Phase 5 — Agents (LangGraph & LangChain) + LangSmith

## Goal
Demonstrate agents that **discover and call MCP tools** and prompts. Show both **LangGraph** and **LangChain** flows. Optional **LangSmith** tracing.

## What you get
- `examples/phase5_langgraph_smoke.py`: 5-step graph
  1) Acquire Entra token
  2) List tools
  3) Call `add`
  4) Read `prynai://status` & `prynai://counter`
  5) Render `quick_summary` and ask OpenAI to summarize
- `examples/phase5_langchain_create_agent_alltools_smoke.py`: build **all** MCP tools into LangChain tools and let the agent pick
- `examples/use_core_langchain_agent.py`: minimal `create_agent` usage
- `src/prynai/mcp_core.py`: **reusable** building block to:
  - fetch token
  - create MCP sessions
  - convert MCP tool catalog → LangChain tools (with Pydantic arg models)
  - (Optionally) return a single “add-only” tool if you want a scoped agent

## Env