mcp-server-demo

Business Request Server

Demo PoC MCP server to be used in other PoC

Devs

Configuration

Before running the server or client, create a .env file from the example:

cp .env.example .env

Edit .env and provide your database credentials and Azure OpenAI settings.

Variable	Description	Required	Default
BITS_DB_SERVER	The hostname or IP address of the SQL Server database.	Yes	missing.domain
BITS_DB_USERNAME	The username for database authentication.	Yes	missing.username
BITS_DB_PWD	The password for database authentication.	Yes	missing.password
BITS_DB_DATABASE	The name of the specific database to connect to.	Yes	missing.dbname
HOST	The host address the MCP server will bind to.	No	0.0.0.0
PORT	The port the MCP server will listen on.	No	8000
MCP_SERVER_URL	The URL where the MCP server is accessible (used by client).	No	http://127.0.0.1:8000/mcp
CORS_ALLOW_ORIGINS	Comma-separated list of origins allowed to bridge to the MCP server.	No	http://localhost,http://127.0.0.1
CORS_ALLOW_CREDENTIALS	Whether to allow credentials (cookies, auth headers) in CORS requests.	No	false
AZURE_OPENAI_ENDPOINT	The endpoint URI for your Azure OpenAI resource (used by client).	Yes (Client)	-
AZURE_OPENAI_VERSION	The API version for Azure OpenAI.	No	2024-05-01-preview
CLIENT_ID	OAuth2 Client ID for authentication (experimental).	No	-
CLIENT_SECRET	OAuth2 Client Secret for authentication (experimental).	No	-
REDIRECT_URI	OAuth2 Redirect URI for authentication (experimental).	No	-

Installation

uv venv
uv pip install -e .
# then run it locally
mcp dev server.py
# or alternatively
python server.py

Navigate to the URL it showed to test your server.

And then you can test functions such as Templates, and then search_business_requests:

Pass in this for query:

{
  "query_filters": [
    {
      "name": "BR_SHORT_TITLE",
      "value": "Server",
      "operator": "="
    }
  ]
}

and for select_fields:

{
  "fields": [
    "BR_SHORT_TITLE"
  ]
}

And then you can filter on the results via filter_results:

[
  {
    "column": "RPT_GC_ORG_NAME_EN",
    "operator": "contains",
    "value": "Correctional"
  }
]

Running via Docker

docker build -t mcp-bits:local .
docker run -p 8080:8080 --env-file ./.env --name mcp-bits-container mcp-bits:local

pymssql issues

pymssql on Mac OSX

pymssql has dependency with FreeTDS, as such ensure you install it beforehand brew install freetds.

After which if you have issues with running the code please do the following:

uv pip uninstall pymssql
uv pip install --pre --no-binary :all: pymssql --no-cache --no-build-isolation

Also you can add to uv pyproject.toml

[tool.uv]
no-binary-package = ["pymssql"]

After this all should be working.

NOTE: Known issue with cython==3.1.0 found here

Here is how to get around it for now (please remove this once this issue is fixed):

uv pip uninstall pymssql
export CFLAGS="-I$(brew --prefix freetds)/include"
export LDFLAGS="-L$(brew --prefix freetds)/lib"
uv pip install "packaging>=24" "setuptools>=54.0" "setuptools_scm[toml]>=8.0" "wheel>=0.36.2" "Cython==3.0.10" "tomli"
uv pip install --pre --no-binary :all: pymssql --no-cache --no-build-isolation

Deployment

CI/CD

TODO

Manual

This is how you can deploy manually in Azure via the CLI.

az webapp deployment source config-local-git \
  --name <WebAppName> \
  --resource-group <ResourceGroupName>
git remote add azure <GitURLFromPreviousStep>
git push azure main

Documentation

• Using this as tutorial on how to build the demo.
• FastMCP documentation
• MCP OAuth 2.0 Authentication