Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/prynai/prynai-mcp

Enterprise-grade MCP server implementation with HTTP streamable transport, OAuth 2.0 authentication, and end-to-end HTTPS encryption. Designed for reproducibility, observability, and secure deployment on Azure Container Apps
https://github.com/prynai/prynai-mcp

agents azurecontainerapps docker mcp oauth2 python redis streamable-http-transport

Last synced: 2 months ago
JSON representation

Enterprise-grade MCP server implementation with HTTP streamable transport, OAuth 2.0 authentication, and end-to-end HTTPS encryption. Designed for reproducibility, observability, and secure deployment on Azure Container Apps

Awesome Lists containing this project

README 

          
# PrynAI MCP — Self-hosted MCP Server



Production-ready **Model Context Protocol (MCP)** server plus client examples that show how to:

- Run locally and in Docker

- Deploy to **Azure Container Apps (ACA)** behind **Entra ID (OAuth2 client credentials)**

- Use **Azure Cache for Redis** for state

- Call MCP tools/resources/prompts from **LangGraph** and **LangChain** agents

- Roll out new server revisions safely (blue/green via ACA revisions)



> 📚 Step-by-step guides live in [`/docs`](./docs): [Phase 1](./docs/phase1.md) · [Phase 2](./docs/phase2.md) · [Phase 3](./docs/phase3.md) · [Phase 4](./docs/phase4.md) · [Phase 5](./docs/phase5.md)



---



## Repo layout



.

├─ server.py # MCP server (tools/resources/prompts)

├─ src/prynai/mcp_core.py # Client helpers: auth, MCP sessions, LangChain tools

├─ examples/ # Smoke tests for each phase

│ ├─ smoke_oauth_ccACADeployment.py

│ ├─ phase5_langgraph_smoke.py

│ ├─ phase5_langchain_create_agent_alltools_smoke.py

│ ├─ use_core_list_tools.py

│ └─ use_core_langchain_agent.py

├─ infra/

│ ├─ docker/Dockerfile

│ ├─ docker-compose.yml

│ └─ azure/

│ ├─ deploy_aca.ps1 # First-time deploy (env + app)

│ └─ deploy_update.ps1 # Remote build in ACR + new ACA revision + traffic

├─ docs/ # Phase guides

└─ .env.example # Sample environment



---



## Features



- ✅ MCP Server (HTTP streamable) with example tools: `add`, `echo`, `long_task`, `bump_counter`, `summarize_via_client_llm`

- ✅ MCP resources (e.g., `prynai://status`, `prynai://counter`)

- ✅ MCP prompts (e.g., `quick_summary`)

- ✅ OAuth2 (Entra ID client credentials) on `/mcp`

- ✅ Redis-backed state (counter)

- ✅ ACA deployment with **remote image builds** in ACR and **multi-revision** rollouts

- ✅ LangGraph + LangChain agents that auto-discover MCP tools



---



## Quick start



### 1) Clone & create env

```bash

git clone https://github.com/PrynAI/PrynAI-MCP.git

cd PrynAI-MCP

cp .env.example .env

# Fill in: PRYNAI_MCP_URL, ENTRA_* , SERVER_APP_ID_URI, OPENAI_API_KEY/OPENAI_MODEL, REDIS_*

```



### 2) Run locally (Python)



```

uv sync

uv run python examples/use_core_list_tools.py

uv run python examples/use_core_langchain_agent.py

```



## Deploy to Azure Container Apps



### First-time (creates env, app, wiring): see docs/phase4.md



### Updates (safe remote build + new revision):



# From repo root

.\infra\azure\deploy_update.ps1

# Script prints preview FQDN; smoke test it, then promote 100% traffic (also printed).



## Versioning & releases



- Tags: MAJOR.MINOR.PATCH-qualifier (e.g., 0.5.0-phase5)



- Create annotated tags and optional GitHub Releases.



- Deployment scripts accept -Tag to pin the image.



## Contributing



- Fork & create a feature branch.



- Add/modify tools in server.py (sync or async are supported).



- Add tests/smoke in examples/.



- Run ruff/black (if configured) and uv run smoke tests.



-Open a PR with a clear description.



⚠️ Never commit secrets. .env is local only.



## Support



- Issues: GitHub Issues



- Security: security@ (responsible disclosure)



---



### `docs/phase1.md`



```markdown

# 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

```



### Run



```

uv run python examples/smoke_http_phase1.py

# Lists tools via HTTP stream to /mcp (no OAuth in this phase)



MCP tools discovered: ['add', 'echo', ...]

```



## Move to Phase 2 to add Redis and stateful behavior.



---



### `docs/phase2.md`



```markdown

# 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

```



### Smoke



# List tools and call counter

uv run python examples/use_core_list_tools.py

# Expect 'bump_counter' and resource 'prynai://counter'



## Troubleshooting



- If counter doesn’t change, verify Redis connectivity and credentials.



- Make sure the server process can reach Redis from its network.



---



### `docs/phase3.md`



```markdown

# 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 `

- 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://`)



Update `.env`:

```



### ENTRA_TENANT_ID=...

### ENTRA_CLIENT_ID=...

### ENTRA_CLIENT_SECRET=...

### SERVER_APP_ID_URI=api://...



## Smoke

```bash

uv run python examples/smoke_oauth_ccACADeployment.py

# Shows 401 if missing token, 200 with token; lists tools & reads resources.

```



## Troubleshooting



- 401/missing_or_malformed: check Authorization header and scopes (/.default).



- SSL issues: ensure you haven’t set custom CA/proxy envs locally unless intended.



---



### `docs/phase4.md`



```markdown

# 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:///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

```



PRYNAI_MCP_URL=https:///mcp

ENTRA_TENANT_ID=...

ENTRA_CLIENT_ID=...

ENTRA_CLIENT_SECRET=...

SERVER_APP_ID_URI=api://...

OPENAI_API_KEY=...

OPENAI_MODEL=gpt-4o-mini



## Optional LangSmith



LANGCHAIN_TRACING_V2=true

LANGSMITH_API_KEY=...

LANGSMITH_PROJECT=PrynAI-MCP



## Run

```bash

uv run python examples/phase5_langgraph_smoke.py

uv run python examples/phase5_langchain_create_agent_alltools_smoke.py

uv run python examples/use_core_langchain_agent.py

```



## Adding new tools



- Implement sync or async functions in server.py. Both work.



- Register them like existing tools; include helpful docstrings (agents read these).



- If args are structured, ensure they’re validated server-side (Pydantic or your schema).



- Re-deploy with infra/azure/deploy_update.ps1 -Tag  and smoke test the new revision before promoting.