MCP服务器部署管理

core-mcp-servers

This repository manages the deployment of MCP servers that don't logically fit within any particular repo.

Each MCP server is deployed as a separate lambda, with API gateway acting as the ingress to them.

Important links

API gateway ingress URLs

Dev gateway
Preprod gateway
Prod gateway

Available MCP servers

Civil Service Acronyms MCP Server

An MCP server for searching the Civil Service Acronym Buster

More information on the website can be found at the Civil Service Acronym Buster github repo

Gov.uk Search MCP Server

An MCP for searching the Gov UK search API for general information from govuk pages.

This provides an up-to-date reference for LLMs to work with when specifically querying for content on gov.uk.

Wikipedia MCP Server

An MCP server for searching Wikipedia and retrieving article content and summaries.

Provides tools to search for pages, get full page content, and summarize topics from Wikipedia.

DEFRA Environment API MCP Server

An MCP server for accessing the DEFRA Environment Agency Public Register.

This server provides comprehensive access to UK environmental data across 11 registries with 22 tools:

General Registry Tools:

• search_across_registries - Search across all registries
• simple_name_search - Simple name/number search

Waste Operations:

• search_waste_operations - Search waste operations registry
• search_for_waste_operation - Get specific waste operation by ID

End of Life Vehicles:

• search_end_of_life_vehicles - Search vehicle facilities registry
• search_for_end_of_life_vehicle - Get specific facility by ID

Enforcement Actions:

• search_enforcement_action - Search enforcement actions registry
• search_for_enforcement_action - Get specific enforcement action by ID

Flood Risk Activity Exemptions:

• search_flood_risk_exemptions - Search flood risk exemptions registry
• search_for_flood_risk_exemption - Get specific exemption by ID

Industrial Installations:

• search_industrial_installations - Search industrial installations registry
• search_for_industrial_installation - Get specific installation by ID

Radioactive Substance Permits:

• search_radioactive_substance - Search radioactive substance permits registry
• search_for_radioactive_substance - Get specific permit by ID

Scrap Metal Dealers:

• search_scrap_metal_dealers - Search scrap metal dealers registry
• search_for_scrap_metal_dealer - Get specific dealer by ID

Waste Exemptions:

• search_waste_exemptions - Search waste exemptions registry
• search_for_waste_exemption - Get specific exemption by ID

Waste Carriers and Brokers:

• search_waste_carriers_brokers - Search waste carriers/brokers registry
• search_for_waste_carrier_broker - Get specific carrier/broker by ID

Water Discharges:

• search_water_discharges - Search water discharge consents registry
• search_for_water_discharge - Get specific consent by ID

Adding a new MCP server

[!WARNING]
Lambda has a maximum unzipped size of 250mb total (code and packages layers), so please keep them short and sweet. If you need more space, please raise this as a new feature request to support containerised MCPs.

Currently, we only support python as an option as getting this to work initially was somewhat convoluted and against spec.

However, we would like to also enable JS lambdas in the future and have that as an option.

To add a new server, the following steps need to be taken:

1. Add another dir to src/ with the name matching the mcp server and lambda you want to deploy
2. Add your code into main.py inside the src/<your_lambda_name>/code/ dir
   1. Use on of the existing lambdas as an example
3. Add required packages into requirements.txt inside the src/<your_lambda_name>/code/ dir
4. Add the deployment and lint steps to the Makefile inside the src/<your_lambda_name>/ dir
5. Add your lambda information to terraform/spec/mcpservers-oas30.yaml file (copy one of the existing ones and just change the references)
6. Add your lambda config as a module call to terraform/lambdas.tf file
7. Update the list of MCP servers to build in the root Makefile, in build_artifacts and lint commands

[!TIP] There are examples in the above terraform files for the terraform steps. And src/gov_uk_acronyms/ for the python part.

Deploying

Deployments will happen automatically when merging to main for prod or when creating a manual release for dev.

To deploy manually using terraform, run make build_artifacts to create the build scripts, and export env=<env>+make tf_apply to deploy.

Security

The api gateway is publicly available on the internet from a network perspective, and has a two layer defence setup.

1. The domain access policy is configured to block all traffic except from known Cabinet Office/DSIT IP addresses.
   1. As this is currently aimed to host tools that are stateless, and use public sources or produce public information, blocking that requires more than this is too restrictive for the time being
   2. Accessing the MCP requires you to know the value of the WAF header token

The intention is to eventually place this behind an auth mechanism of some kind that matches the MCP spec, if and when the hosted tools demand it.