feat: provide the chat of the rag by MCP (#27)

This pull request introduces a new component, the MCP Server, to the RAG system. The MCP Server facilitates interaction between MCP-compatible clients and the RAG backend, providing features like simple chat interfaces, conversation history, and citation support. The changes include updates to documentation, build configurations, and the addition of new files and dependencies for the MCP Server.

### MCP Server Introduction and Features
* **New Component Documentation**: Added a detailed description of the MCP Server in `README.md` and created a dedicated `mcp-server/README.md` to explain its features, architecture, configuration, and deployment. [[1]](diffhunk://#diff-b335630551682c19a781afebcf4d07bf978fb1f8ac04c6bf87428ed5106870f5R50) [[2]](diffhunk://#diff-de8068a21895dfaa62ca644f5c42338393947c7b760c8862c7ec34a2abb29bf1R1-R106)
* **Dockerfile for MCP Server**: Added `mcp-server/Dockerfile` to define the containerized environment for the MCP Server, including dependency installation and runtime setup.

### Build and Deployment Configuration
* **Tiltfile Updates**: Integrated the MCP Server into the development workflow with live updates, linting, and port-forwarding configurations in `Tiltfile`. Enabled the MCP feature in Helm values. [[1]](diffhunk://#diff-c2ee8653e1d6b85f0aadf87cd438a9250806c052877248442be4d434cbc52425R180-R210) [[2]](diffhunk://#diff-c2ee8653e1d6b85f0aadf87cd438a9250806c052877248442be4d434cbc52425R345-R346) [[3]](diffhunk://#diff-c2ee8653e1d6b85f0aadf87cd438a9250806c052877248442be4d434cbc52425R432-R436)
* **Makefile Update**: Added a build command for the MCP Server to the `Makefile`.

### Codebase Additions
* **Dependency Injection**: Introduced `mcp-server/src/dependency_container.py` for managing application dependencies using `dependency-injector`.
* **Main Entry Point**: Added `mcp-server/src/main.py` to initialize and run the MCP Server.
* **OpenAPI Client**: Generated OpenAPI client files under `mcp-server/src/rag_backend_client/openapi_client/` for backend communication. [[1]](diffhunk://#diff-08494b7fe5b52ed37ca18f372ed78c8e17b6a13fcea8c33ee860af361fb7b0ffR1-R44) [[2]](diffhunk://#diff-36e13e6a9f7a6f22ba251ad8f60039b94d194a20ce9c3f6f249feed038d61ea2R1-R4) [[3]](diffhunk://#diff-e9ff3d7051093df73131df1a7e1201884e1ef2d0d6673573e59acaf32f4e1d7cR1-R20) [[4]](diffhunk://#diff-c7187b1f33a4416b9df23604777a695bf0d52aba4dbc2dcbecbf834cf446fceeR1-R30)

### Supporting Files and Dependencies
* **Project Configuration**: Added `mcp-server/pyproject.toml` to define dependencies and development tools for the MCP Server.
* **OpenAPI Generator Ignore**: Added `.openapi-generator-ignore` to exclude generated files from being overwritten.
* **API Generator Script**: Added `mcp-server/api-generator.sh` to automate the generation of OpenAPI client code.

### Miscellaneous
* **Dockerignore Update**: Updated `.dockerignore` to exclude unnecessary files for the MCP Server.

Author: a-klos

.dockerignore

@@ -18,3 +18,6 @@ frontend/*.log
 */.venv
 */.env
 */*.pyc
+
+
+**/.notebooks

.gitignore

@@ -12,7 +12,9 @@ auth
 node_modules/
 
 **/.notebooks
+**/.notebooks/*
 **/todo*.md
+**/mcp.json
 
 # Byte-compiled / optimized / DLL files
 __pycache__/

Makefile

@@ -27,3 +27,4 @@ build_and_push:
 	docker buildx build --platform linux/amd64 -t $(REGISTRY)/document-extractor:$(IMAGE_TAG)  -f document-extractor/Dockerfile --push .
 	docker buildx build --platform linux/amd64 -t $(REGISTRY)/frontend:$(IMAGE_TAG)  -f frontend/apps/chat-app/Dockerfile --push .
 	docker buildx build --platform linux/amd64 -t $(REGISTRY)/admin-frontend:$(IMAGE_TAG)  -f frontend/apps/admin-app/Dockerfile --push .
+    docker buildx build --platform linux/amd64 -t $(REGISTRY)/mcp-server:$(IMAGE_TAG)  -f mcp-server/Dockerfile --push .

README.md

@@ -47,6 +47,7 @@ This repository contains the following components:
 - [*rag-backend*](#111-rag-backend): The main component of the RAG.
 - [*admin-backend*](#112-admin-backend): Manages user documents and confluence spaces, interacts with document-extractor and rag-backend.
 - [*document-extractor*](#113-document-extractor): Extracts content from documents and Confluence spaces.
+- [*mcp-server*](#114-mcp-server): Model Context Protocol server that provides MCP-compatible access to the RAG system.
 - *frontend*: Frontend for both, chat and admin APIs.
 - *rag-infrastructure*: Contains the helm-chart and other files related to infrastructure and deployment. Please consult [this README](https://github.com/stackitcloud/rag-infrastructure/blob/main/README.md) for further information.
 - *rag-core-library*: Contains the API-libraries that are used to construct the backend-services in this repository. For further information, please consult [this README](https://github.com/stackitcloud/rag-core-library/blob/main/README.md).
@@ -68,6 +69,17 @@ The Document extractor is a component that is used to extract the content from t
 
 All components are provided by the *extractor-api-lib*. For further information on endpoints and requirements, please consult [this README](https://github.com/stackitcloud/rag-core-library/blob/main/README.md#3-extractor-api-lib).
 
+#### 1.1.4 MCP Server
+
+The MCP Server is a Model Context Protocol (MCP) server that provides a bridge between MCP-compatible clients and the RAG backend. It enables AI assistants and other tools to interact with the RAG system through standardized MCP tools.
+
+The MCP server runs as a sidecar container alongside the main RAG backend and exposes two main tools:
+
+- `chat_simple`: Basic question-answering without conversation history
+- `chat_with_history`: Advanced chat interface with conversation history and returns structured responses with `answer`, `finish_reason`, and `citations`.
+
+For further information on configuration and usage, please consult the [MCP Server README](./mcp-server/README.md).
+
 ### 1.2 Requirements
 
 > 📝  *Windows users*: make sure you use wsl for infrastructure setup & orchestration.

Tiltfile

@@ -177,8 +177,37 @@ local_resource(
     allow_parallel=True,
 )
 
+################################## build mcp image and do live update ##################################################
+# NOTE: full image names should match the one in the helm chart values.yaml!
+registry = "ghcr.io/stackitcloud/rag-template"
+mcp_image_name = "rag-mcp"
+
+mcp_context = "./mcp-server"
+mcp_full_image_name = "%s/%s" % (registry, mcp_image_name)
+docker_build(
+    mcp_full_image_name,
+    ".",
+    build_args={
+        "dev": "1" if backend_debug else "0",
+    },
+    live_update=[
+        sync(mcp_context, "/app/mcp-server"),
+    ],
+    dockerfile=mcp_context + "/Dockerfile",
+)
+
+# Add linter trigger
+local_resource(
+    "MCP server linting",
+    create_linter_command(mcp_context, "back"),
+    labels=["linting"],
+    auto_init=False,
+    trigger_mode=TRIGGER_MODE_AUTO,
+    allow_parallel=True,
+)
+
 ########################################################################################################################
-################################## build admin backend image and do live update ##############################################
+################################## build admin backend image and do live update ########################################
 ########################################################################################################################
 
 # NOTE: full image names should match the one in the helm chart values.yaml!
@@ -313,6 +342,8 @@ value_override = [
     "features.minio.enabled=true",
     "shared.config.tls.enabled=false",
     "shared.ssl=false",
+    "shared.config.basicAuth.enabled=true",
+    "features.mcp.enabled=true",
     # ingress host names
     "backend.ingress.host.name=rag.localhost",
     # langfuse
@@ -398,6 +429,11 @@ k8s_resource(
             31415,
             container_port=31415,
             name="Backend-Debugger",
+        ),
+        port_forward(
+            9090,
+            container_port=8000,
+            name="MCP-Server",
         )
     ],
     labels=["backend"],

mcp-server/Dockerfile

@@ -0,0 +1,62 @@
+FROM --platform=linux/amd64 python:3.11.7-bookworm AS build
+
+ARG dev=0
+ENV POETRY_VIRTUALENVS_PATH=/app/mcp-server/.venv
+ENV POETRY_VERSION=1.8.3
+
+WORKDIR /app
+
+RUN DEBIAN_FRONTEND=noninteractive apt-get update \
+    && DEBIAN_FRONTEND=noninteractive apt-get install -y build-essential --no-install-recommends make && \
+    python3 -m venv "${POETRY_VIRTUALENVS_PATH}" \
+    && $POETRY_VIRTUALENVS_PATH/bin/pip install "poetry==${POETRY_VERSION}"
+ENV PATH="${POETRY_VIRTUALENVS_PATH}/bin:$PATH"
+
+
+WORKDIR /app/mcp-server
+COPY mcp-server/pyproject.toml mcp-server/poetry.lock ./
+
+RUN mkdir log && chmod 700 log
+RUN touch /app/mcp-server/log/logfile.log && chmod 600 /app/mcp-server/log/logfile.log
+
+RUN poetry config virtualenvs.create false &&\
+    if [ "$dev" = "1" ]; then \
+    poetry install --no-interaction --no-ansi --no-root --with dev; \
+    else \
+    poetry install --no-interaction --no-ansi --no-root; \
+    fi
+
+FROM --platform=linux/amd64 python:3.11.7-bookworm
+ARG dev=0
+
+RUN adduser --disabled-password --gecos "" --uid 65532 nonroot
+
+ENV POETRY_VIRTUALENVS_PATH=/app/mcp-server/.venv
+COPY --from=build --chown=nonroot:nonroot ${POETRY_VIRTUALENVS_PATH} ${POETRY_VIRTUALENVS_PATH}
+COPY --from=build /usr/local/bin/ /usr/local/bin/
+COPY --from=build /usr/bin/make /usr/bin/make
+COPY --from=build /usr/local/lib/ /usr/local/lib/
+
+WORKDIR /app/mcp-server
+COPY --chown=nonroot:nonroot mcp-server/src ./src
+COPY --chown=nonroot:nonroot mcp-server/pyproject.toml mcp-server/poetry.lock ./
+# cleanup
+RUN apt-get clean autoclean
+RUN apt-get autoremove --yes
+
+RUN if [ "$dev" = "0" ]; then \
+        while read -r shell; do rm -f "$shell"; done < /etc/shells; \
+        rm -rf /var/lib/{apt,dpkg,cache,log}/ \
+    else \
+        echo "POETRY_VIRTUALENVS_PATH=/app/mcp-server/.venv" >> /etc/environment;\
+        export POETRY_VIRTUALENVS_PATH=/app/mcp-server/.venv;\
+        export PATH="${POETRY_VIRTUALENVS_PATH}/bin:$PATH";\
+    fi
+
+
+USER nonroot
+COPY --from=build --chown=nonroot:nonroot  /app/mcp-server/log /app/mcp-server/log
+
+ENV PATH="${POETRY_VIRTUALENVS_PATH}/bin:${PATH}"
+
+CMD [ "poetry", "run", "python", "src/main.py" ]

mcp-server/README.md

@@ -0,0 +1,106 @@
+# MCP Server
+
+The MCP Server is a Model Context Protocol (MCP) server that provides a bridge between MCP-compatible clients and the RAG backend. It enables AI assistants and other tools to interact with the RAG system through standardized MCP tools.
+
+## Features 🚀
+
+- **Simple Chat Interface**: Basic question-answering without conversation history
+- **Chat with History**: Conversational interface that maintains context across messages
+- **Citation Support**: Returns source documents and metadata for transparency
+- **Streamable HTTP Transport**: Uses HTTP-based transport for reliable communication
+- **Configurable Settings**: Environment-based configuration for different deployment scenarios
+
+## Architecture
+
+The server consists of several key components:
+
+- **RagMcpServer**: Main server class that handles MCP tool registration and request routing
+- **Dependency Container**: Manages dependency injection for clean architecture
+- **Settings**: Environment-based configuration management
+- **RAG Backend Client**: Auto-generated OpenAPI client for backend communication
+
+## Requirements
+
+All required python libraries can be found in the [pyproject.toml](pyproject.toml) file.
+The MCP server uses Poetry for dependency management and shares the base Dockerfile pattern with other services in the RAG template.
+
+## Available Tools
+
+The server exposes two main MCP tools for interacting with the RAG system:
+
+### `chat_simple`
+
+Simple question-answering interface that returns plain text responses.
+
+**Parameters:**
+
+- `session_id` (str): Unique identifier for the chat session
+- `message` (str): The question or message to send to the RAG system
+
+**Returns:**
+
+- `str`: Plain text answer from the RAG system
+
+### `chat_with_history`
+
+Advanced chat interface that supports conversation history and returns structured responses with citations.
+
+**Parameters:**
+
+- `session_id` (str): Unique identifier for the chat session
+- `message` (str): The current question or message
+- `history` (list[dict], optional): Previous conversation history
+
+**History Format:**
+Each history item should be a dictionary with:
+
+- `role`: Either "user" or "assistant"
+- `message`: The message content
+
+**Returns:**
+
+- `dict`: Structured response containing:
+  - `answer`: The response text
+  - `finish_reason`: Why the response ended
+  - `citations`: List of source documents with content and metadata
+
+## Configuration
+
+The server supports configuration through environment variables with the following prefixes:
+
+### MCP Settings (`MCP_` prefix)
+
+- `MCP_HOST`: Server bind address (default: `0.0.0.0`)
+- `MCP_PORT`: Server port (default: `8000`)
+- `MCP_NAME`: Server name (default: `RAG MCP server`)
+
+### Backend Settings (`BACKEND_` prefix)
+
+- `BACKEND_BASE_PATH`: RAG backend URL (default: `http://127.0.0.1:8080`)
+
+## Deployment
+
+The MCP server is designed to be deployed alongside the main RAG backend as a sidecar container. A detailed explanation of the deployment can be found in the [main README](../README.md) and the [infrastructure README](../rag-infrastructure/README.md) of the project.
+
+### Docker Support
+
+The server includes Docker support for containerized deployment and is integrated into the main Tilt development workflow.
+
+### Integration in RAG Template
+
+The MCP server is automatically deployed when `backend.mcp.enabled=true` is set in the Helm values. It runs as a sidecar container alongside the main RAG backend, accessible via:
+
+- **Port**: 8000 (configurable via `MCP_PORT`)
+- **Endpoint**: `/mcp` path through the main ingress
+- **Development**: Port-forwarded to 9090 in local Tilt setup
+
+## Development
+
+The MCP server is integrated into the main RAG template development workflow:
+
+- **Tilt Integration**: Automatically built and deployed with live reload
+- **Linting**: Included in the main linting pipeline
+- **Testing**: Part of the overall test suite
+- **Debugging**: Supports the same debugging workflow as other services
+
+For detailed development setup instructions, see the [main README](../README.md).

mcp-server/api-generator.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+        
+docker run --user $(id -u):$(id -g) --rm -v $PWD:/local openapitools/openapi-generator-cli@sha256:b35aee2d0f6ffadadcdad9d8fc3c46e8d48360c20b5731a5f47c809d51f67a04 generate -i /local/rag-core-library/rag-core-api/openapi.yaml -g python -o /local/mcp-server/src --additional-properties=generateSourceCodeOnly=True,packageName=rag_backend_client.openapi_client
+cd ./mcp-server
+black .
+cd ..