Enkrypt AI Secure MCP Gateway

Overview

This Secure MCP Gateway is built with authentication, automatic tool discovery, caching, and guardrail enforcement.

It sits between your MCP client and MCP servers. So, by it's nature it itself also acts as an MCP server as well as an MCP client :)

When your MCP client connects to the Gateway, it acts as an MCP server. When the Gateway connects to the actual MCP server, it acts as an MCP client.

• Pypi Package

• Docker Image

• Also see:

  • CLI-Commands-Reference.md for the list of commands and their usage

  • API-Reference.md for the list of API endpoints and their usage

  • MCP Gateway Setup Notebook for a complete walkthrough of all the essential commands

Table of Contents

• 1. Features 🚀

• 2. High level steps of how the MCP Gateway works 🪜

• 3. Prerequisites 🧩

• 4. Gateway Setup 👨‍💻

• 5. (Optional) OpenTelemetry Setup 📊

• 6. Verify Installation and check the files generated ✅

• 7. Edit the Gateway config as needed ✏️

• 8. (Optional) Add GitHub MCP Server to the Gateway 🤖

• 9. (Optional) Protect GitHub MCP Server and Test Echo Server 🔒

• 10. Recommendations for using Guardrails 💡

• 11. Other tools available 🔧

• 12. Deployment Patterns 🪂

• 13. Uninstall the Gateway 🗑️

• 14. Troubleshooting 🕵

• 15. Known Issues being worked on 🏗️

• 16. Known Limitations ⚠️

• 17. Contribute 🤝

1. Features

Below are the list of features Enkrypt AI Secure MCP Gateway provides:

1. Authentication: We use Unique Key to authenticate with the Gateway. We also use Enkrypt API Key if you want to protect your MCPs with Enkrypt Guardrails

2. Ease of use: You can configure all your MCP servers locally in the config file or better yet in Enkrypt (Coming soon) and use them in the Gateway by using their name

3. Dynamic Tool Discovery: The Gateway discovers tools from the MCP servers dynamically and makes them available to the MCP client

4. Restrict Tool Invocation: If you don't want all tools to be accessible of a an MCP server, you can restrict them by explicitly mentioning the tools in the Gateway config so that only the allowed tools are accessible to the MCP client

5. Caching: We cache the user gateway config and tools discovered from various MCP servers locally or in an external cache server like KeyDB if configured to improve performance

6. Guardrails: You can configure guardrails for each MCP server in Enkrypt both on input side (before sending the request to the MCP server) and output side (after receiving the response from the MCP server)

7. Logging: We log every request and response from the Gateway locally in your MCP logs and also forward them to Enkrypt (Coming soon) for monitoring. This enables you to see all the calls made in your account, servers used, tools invoked, requests blocked, etc.

1.1 Guardrails

Input Protection: Topic detection, NSFW filtering, toxicity detection, injection attack prevention, keyword detection, policy violation detection, bias detection, and PII redaction (More coming soon like system prompt protection, copyright protection, etc.)

Output Protection: All input protections plus adherence checking and relevancy validation (More coming soon like hallucination detection, etc.) We also auto unredact the response if it was redacted on input.

1.2 Concepts

• MCP Config is an array of MCP servers like mcp_server_1, mcp_server_2, mcp_server_3 etc.

  • Each config has a unique ID

• User is a user of the gateway with unique email and ID

• A project is a collection of users that share an MCP Config

  • Project has a name and unique ID

  • The MCP Config can be updated or can be pointed to a different config by the Admin

  • Users can be added to multiple projects

• An API Key is created for a user and project combination

  • A user can have different API Keys for different projects

  • This API Key is used to authenticate the user and identify the right project and MCP Config

• See

2. High level steps of how the MCP Gateway works

1. Your MCP client connects to the Secure MCP Gateway server with API Key (handled by src/secure_mcp_gateway/gateway.py).

2. Gateway server fetches gateway config from local enkrypt_mcp_config.json file or remote Enkrypt Auth server (Coming soon).

   • It caches the config locally or in an external cache server like KeyDB if configured to improve performance.

3. If input guardrails are enabled, request is validated before the tool call (handled by src/secure_mcp_gateway/guardrail.py).

   • Request is blocked if it violates any of the configured guardrails and the specific detector is configured to block.

4. Requests are forwarded to the Gateway Client (handled by src/secure_mcp_gateway/client.py).

5. The Gateway client forwards the request to the appropriate MCP server (handled by src/secure_mcp_gateway/client.py).

6. The MCP server processes the request and returns the response to the Gateway client.

7. If it was a discover tools call, the Gateway client caches the tools locally or in an external cache server like KeyDB if configured. It then forwards the response to the Gateway server.

8. The Gateway server receives the response from the Gateway client and if output guardrails are enabled, it validates the response against the configured guardrails (handled by src/secure_mcp_gateway/guardrail.py).

   • Response is blocked if it violates any of the configured guardrails and the specific detector is configured to block.

9. The Gateway server forwards the response back to the MCP client if everything is fine.

3. Prerequisites

• Git 2.43 or higher

• Python 3.11 or higher installed on your system and is accessible from the command line using either python or python3 command

• pip 25.0.1 or higher is installed on your system and is accessible from the command line using either pip or python -m pip command

• uv 0.7.9 or higher is installed on your system and is accessible from the command line using either uv or python -m uv command

• Check if Python, pip and uv are installed

• If any of the below commands fail, please refer the respective documentation to install them properly

• Install Claude Desktop as the MCP Client from their website if you haven't already and login to it

  • If you are using Linux and cannot run any

• Any other dependencies required for the MCP servers we want to proxy requests to

  • Follow the instructions of the respective MCP server to install its dependencies

  • Like Node.js, npx, docker, etc.

• (Optional) A cache server like KeyDB installed and running (If you want to cache externally and not locally)

If you want to protect your MCPs with Enkrypt Guardrails, you need to do the following:

• Create a new account if you don't have one. It's free! 🆓 No credit card required 💳🚫

• An ENKRYPT_API_KEY which you can get from Enkrypt Dashboard Settings

• To protect your MCPs with Guardrails, you can use the default sample Guardrail Sample Airline Guardrail to get started or you can create your own custom Guardrail

• To configure custom Guardrails, you need to either login to Enkrypt AI App or use the APIs/SDK

  • Create Guardrails in Enkrypt AI App Dashboard ✅

  • Create Guardrails using APIs

  • Create Guardrails using SDK

  • You can also use Enkrypt MCP Server 🤯 to create Guardrails and use them in the Gateway

4. Gateway Setup

4.1 Local Installation with pip

4.1.1 Download and Install the Package

• Activate a virtual environment

  python -m venv .secure-mcp-gateway-venv # Activate the virtual environment # On Windows .secure-mcp-gateway-venv\Scripts\activate # On Linux/macOS source .secure-mcp-gateway-venv/bin/activate # Run the below to exit the virtual environment later if needed deactivate

• Install the package. For more info see https://pypi.org/project/secure-mcp-gateway/

  pip install secure-mcp-gateway

4.1.2 Run the Generate Command

• This generates the config file at

  secure-mcp-gateway generate-config

4.1.3 Example of the generated config file

• This is an example of the default configuration file generated by the CLI on macOS:

• This is an example of the default configuration file generated by the CLI on Windows:

4.1.4 Install the Gateway for Claude Desktop

• Run the following command to install the gateway for Claude:

  secure-mcp-gateway install --client claude-desktop

• This will register Enkrypt Secure MCP Gateway with Claude Desktop.

• NOTE: Please restart Claude Desktop after installation

4.1.5 Example of the Claude Desktop Config after installation

• ~/Library/Application Support/Claude/claude_desktop_config.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "/opt/homebrew/bin/uv", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "/Users/user/enkryptai/secure-mcp-gateway/venv/lib/python3.13/site-packages/secure_mcp_gateway/gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

• %USERPROFILE%\AppData\Roaming\Claude\claude_desktop_config.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "C:\\Users\\<User>\\.local\\bin\\uv.EXE", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "C:\\Users\\<User>\\Documents\\GitHub\\EnkryptAI\\secure-mcp-gateway\\.secure-mcp-gateway-venv\\Lib\\site-packages\\secure_mcp_gateway\\gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

4.1.6 Install the Gateway for Cursor

• Run the CLI Install Command for Cursor

  secure-mcp-gateway install --client cursor

• This automatically updates your ~/.cursor/mcp.json (on Windows it is at: %USERPROFILE%.cursor\mcp.json) with the correct entry.

• Although it is not usually required to restart, if you see it in loading state for a long time, please restart Cursor

• ~/.cursor/mcp.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "uv", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "/Users/user/enkryptai/secure-mcp-gateway/venv/lib/python3.13/site-packages/secure_mcp_gateway/gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

• %USERPROFILE%\.cursor\mcp.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "uv", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "C:\\Users\\<User>\\Documents\\GitHub\\EnkryptAI\\secure-mcp-gateway\\.secure-mcp-gateway-venv\\Lib\\site-packages\\secure_mcp_gateway\\gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

4.2 Local Installation with git clone

4.2.1 Clone the repo, setup virtual environment and install dependencies

• Clone the repository:

• Install Python dependencies:

• Verify mcp cli got installed successfully:

4.2.2 Run the setup script

• This script creates the config file at ~/.enkrypt/enkrypt_mcp_config.json on macOS and %USERPROFILE%\.enkrypt\enkrypt_mcp_config.json on Windows based on src/secure_mcp_gateway/example_enkrypt_mcp_config.json file

• It replaces UNIQUE_GATEWAY_KEY and other UUIDs with auto generated values and also replaces DUMMY_MCP_FILE_PATH with the actual path to the test MCP file test_mcps/echo_mcp.py

• It also installs the MCP client in Claude Desktop

• NOTE: Please restart Claude Desktop after running the setup script to see the Gateway running in Claude Desktop

4.2.3 Setup Other MCP Clients

• You can navigate to cursor's Global MCP file at ~/.cursor/mcp.json on Linux/macOS or %USERPROFILE%\.cursor\mcp.json on Windows

  • If you would like to use at a Project level place it inside your project. For details see Cursor's docs

• You can also navigate to the file Via cursor's UI by clicking on settings gear icon on the top right

  

• Click on MCP and then click on Add new global MCP server which takes you to the mcp.json file

  

• Example mcp.json file opened in the editor

  

• Once the file is opened at Global or Project level, you can copy paste the same config we used in Claude Desktop. For reference, you can refer to Installation - 5.2 Example MCP config file generated 📄

  • Be sure to use your own file that was generated by the

• See Verify Cursor section to verify the MCP server is running in Cursor

4.3 Docker Installation

4.3.1 Build the Docker Image

4.3.2 Generate the config file

• This creates a config file in the ~/.enkrypt/docker/enkrypt_mcp_config.json file on macOS/Linux and %USERPROFILE%\.enkrypt\docker\enkrypt_mcp_config.json file on Windows.

4.3.3 Install the Gateway in Claude Desktop

• You can find the Claude config location at the below locations in your system. For reference see Claude docs.

  • macOS: ~/Library/Application Support/Claude

  • Windows: %APPDATA%\Claude

4.3.4 Example Claude Desktop config file

4.3.5 Install the Gateway in Cursor

• You can find the Cursor config location at the below locations. For reference see Cursor docs.

  • macOS: ~/.cursor

  • Windows: %USERPROFILE%\.cursor

4.4 Remote Installation

4.4.1 Run the Gateway in a remote server

• Or run in k8s using our docker image enkryptai/secure-mcp-gateway:vx.x.x

• Example: enkryptai/secure-mcp-gateway:v2.1.0

• Use the latest version from Docker Hub: https://hub.docker.com/r/enkryptai/secure-mcp-gateway/tags

• You can either mount the config file locally or download the json file from a remote place like S3 using an initContainer and mount the volume

• See docs/secure-mcp-gateway-manifest-example.yaml for the complete manifest file reference

4.4.2 Modify your MCP Client config to use the Gateway

• You can find the Claude config location at the below locations in your system. For reference see Claude docs.

  • macOS: ~/Library/Application Support/Claude

  • Windows: %APPDATA%\Claude

• You can find the Cursor config location at the below locations. For reference see Cursor docs.

  • macOS: ~/.cursor

  • Windows: %USERPROFILE%\.cursor

• Replace the ENKRYPT_GATEWAY_KEY with the key you got from the enkrypt_mcp_config.json file

• Replace the http://0.0.0.0:8000/mcp/ with the http(s)://<remote_server_ip>:<port>/mcp/

• If you are running this locally, you can use http://0.0.0.0:8000/mcp/

• You can setup ingress to route the traffic to the MCP Gateway over https

• Example: https://mcp.enkryptai.com/mcp/

• NOTE: Please make sure node and npm are installed on the client machine

  • To verify, run node -v and npm -v

• NOTE: Make sure to use the trailing slash

5. (Optional) OpenTelemetry Setup

This section explains how to set up and use OpenTelemetry (OTEL) with the Enkrypt Secure MCP Gateway for observability.

5.1 Architecture

The observability stack includes:

• OpenTelemetry Collector: Collects telemetry data (traces, metrics, logs)

• Jaeger: Distributed tracing visualization

• Loki: Log aggregation and querying

• Prometheus: Metrics aggregation

• Grafana: Unified visualization for metrics and logs

  • Traces are not visible in Grafana for some reason. Please use Jaeger for traces.

5.2 Prerequisites

• Docker and Docker Compose installed

• Gateway installed and running (follow section 4)

5.3 Setup Steps

1. Start the Observability Stack

   cd infra docker-compose up -d

2. To stop the Observability Stack

   # When we want to stop the Observability Stack, run the below command docker-compose down

5.4 Configuration

• Edit the enkrypt_mcp_config.json file to enable telemetry

  { "common_mcp_gateway_config": { ... "enkrypt_telemetry": { "enabled": true, "insecure": true, "endpoint": "http://localhost:4317" } }, ... }

5.5 Verification Steps

1. Verify Services are Running

   # On Windows docker ps | findstr "loki grafana jaeger otel prometheus" # On Linux/macOS docker ps | grep -E "loki|grafana|jaeger|otel|prometheus"

2. Access Service UIs

   • Grafana: http://localhost:3000 (default credentials: admin/admin)

   • Jaeger: http://localhost:16686

   • Prometheus: http://localhost:9090

   • Loki: Access through Grafana

     1. Open Grafana (http://localhost:3000)

     2. Go to Explore (left sidebar)

     3. Select "Loki" from the data source dropdown

3. Verify Gateway Telemetry

   • Make test requests through the Gateway like List all servers and tools and echo test

   • Check traces in Jaeger:

     • Add optional tags like enkrypt_email=default@example.com or enkrypt_project_name=default_project or enkrypt_mcp_config_id=fcbd4508-1432-4f13-abb9-c495c946f638 to see the traces for a specific user, project or MCP config etc.

     • We can also combine tags by separating them with spaces like enkrypt_email=default@example.com enkrypt_project_name=default_project

     • Look for enkrypt_discover_all_tools spans

     • Examine child spans for cache, tool discovery, etc.

       

   • Check metrics in Grafana:

     • Navigate to Drilldown -> metrics

     • We can filter on various labels like email, user_id, mcp_config_id, project_id, project_name etc.

       

   • Check logs in Grafana

     • Navigate to Drilldown -> Logs

     • Select label as service_name=secure-mcp-gateway and click Show logs

     • Now we can filter by various labels like attributes_project_name, attributes_project_id, attributes_email, attributes_user_id, attributes_mcp_config_id, attributes_tool_name etc.

       

   • Check Dashboards in Grafana by navigating to Dashboards -> OpenTelemetry Gateway Metrics

     • Due to issues in Grafana, you may need to edit each tile and click

5.6 Available Telemetry (Not exhaustive)

1. Traces

   • Request processing pipeline

   • Tool invocations with duration tracking

   • Cache operations (hits/misses)

   • Guardrail checks

   • Error tracking and status monitoring

   • Detailed attributes for debugging

2. Metrics

   • enkrypt_list_all_servers_calls: API endpoint usage

   • mcp_cache_misses_total: Cache efficiency tracking

   • enkrypt_servers_discovered: Server discovery monitoring

   • mcp_tool_calls_total: Tool invocation tracking

   • mcp_tool_call_duration_seconds: Performance monitoring (histogram)

3. Logs

   • Structured JSON format for better querying

   • Gateway operations with context

   • Error conditions with stack traces

   • Security events and guardrail checks

   • Performance data with timing information

6. Verify Installation and check the files generated

6.1 Verify Claude Desktop

• To verify Claude installation, navigate to claude_desktop_config.json file by following these instructions

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

6.2 Example MCP config file generated

• ~/Library/Application Support/Claude/claude_desktop_config.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "/opt/homebrew/bin/uv", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "/Users/user/enkryptai/secure-mcp-gateway/src/secure_mcp_gateway/gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

• %USERPROFILE%\AppData\Roaming\Claude\claude_desktop_config.json

  { "mcpServers": { "Enkrypt Secure MCP Gateway": { "command": "C:\\Users\\<User>\\.local\\bin\\uv.EXE", "args": [ "run", "--with", "mcp[cli]", "mcp", "run", "C:\\Users\\<User>\\Documents\\GitHub\\EnkryptAI\\secure-mcp-gateway\\src\\secure_mcp_gateway\\gateway.py" ], "env": { "ENKRYPT_GATEWAY_KEY": "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat", "ENKRYPT_PROJECT_ID": "3c09f06c-1f0d-4153-9ac5-366397937641", "ENKRYPT_USER_ID": "6469a670-1d64-4da5-b2b3-790de21ac726" } } } }

6.3 Restart Claude Desktop to run the Gateway

• After restarting, navigate to Claude Desktop Settings

  

• Click on Developer -> Enkrypt Secure MCP Gateway

  

• You can also click on the settings icon below the search bar to see the Gateway in available

  

• Click on Enkrypt Secure MCP Gateway to see the list of tools available

  

• You can check Claude logs while asking Claude to do something to see the Gateway in action

  • Example 🍎 Linux/macOS log path: ~/Library/Application Support/Claude/logs/mcp-server-Enkrypt Secure MCP Gateway.log

  • Example 🪟 Windows log path: %USERPROFILE%\AppData\Roaming\Claude\logs\mcp-server-Enkrypt Secure MCP Gateway.log

6.4 Example prompts

• list all servers, get all tools available and echo test

  • This uses a test MCP server echo_server which is in test_mcps/echo_mcp.py

• We can also combine multiple prompts into one that trigger multiple tool calls at once

• Example: echo test and also echo best

• Example:

• This could be a malicious prompt but because no guardrails are enabled, it will not be blocked

6.5 Example config file generated

• Example enkrypt_mcp_config.json generated by the setup script in ~/.enkrypt/enkrypt_mcp_config.json on macOS and %USERPROFILE%\.enkrypt\enkrypt_mcp_config.json on Windows

• If you ran docker command to install the Gateway, the config file will be in

  { "common_mcp_gateway_config": { "enkrypt_log_level": "INFO", "enkrypt_base_url": "https://api.enkryptai.com", "enkrypt_api_key": "YOUR_ENKRYPT_API_KEY", "enkrypt_use_remote_mcp_config": false, "enkrypt_remote_mcp_gateway_name": "enkrypt-secure-mcp-gateway-1", "enkrypt_remote_mcp_gateway_version": "v1", "enkrypt_mcp_use_external_cache": false, "enkrypt_cache_host": "localhost", "enkrypt_cache_port": 6379, "enkrypt_cache_db": 0, "enkrypt_cache_password": null, "enkrypt_tool_cache_expiration": 4, "enkrypt_gateway_cache_expiration": 24, "enkrypt_async_input_guardrails_enabled": false, "enkrypt_async_output_guardrails_enabled": false, "enkrypt_telemetry": { "enabled": true, "insecure": true, "endpoint": "http://localhost:4317" } }, "mcp_configs": { "fcbd4508-1432-4f13-abb9-c495c946f638": { "mcp_config_name": "default_config", "mcp_config": [ { "server_name": "echo_server", "description": "Simple Echo Server", "config": { "command": "python", "args": [ "C:\\Users\\<User>\\Documents\\GitHub\\EnkryptAI\\secure-mcp-gateway\\src\\secure_mcp_gateway\\test_mcps\\echo_mcp.py" ] }, "tools": {}, "input_guardrails_policy": { "enabled": false, "policy_name": "Sample Airline Guardrail", "additional_config": { "pii_redaction": false }, "block": [ "policy_violation" ] }, "output_guardrails_policy": { "enabled": false, "policy_name": "Sample Airline Guardrail", "additional_config": { "relevancy": false, "hallucination": false, "adherence": false }, "block": [ "policy_violation" ] } } ] } }, "projects": { "3c09f06c-1f0d-4153-9ac5-366397937641": { "project_name": "default_project", "mcp_config_id": "fcbd4508-1432-4f13-abb9-c495c946f638", "users": [ "6469a670-1d64-4da5-b2b3-790de21ac726" ], "created_at": "2025-07-16T17:02:00.406877" } }, "users": { "6469a670-1d64-4da5-b2b3-790de21ac726": { "email": "default@example.com", "created_at": "2025-07-16T17:02:00.406902" } }, "apikeys": { "2W8UupCkazk4SsOcSu_1hAbiOgPdv0g-nN9NtfZyg-rvYGat": { "project_id": "3c09f06c-1f0d-4153-9ac5-366397937641", "user_id": "6469a670-1d64-4da5-b2b3-790de21ac726", "created_at": "2025-07-16T17:02:00.406905" } } }

6.6 Verify Cursor

• You can see the MCP server in the list of MCP servers in Cursor by navigating to ~/.cursor/mcp.json and also by clicking on the settings icon on the top right and then clicking on Tools & Integrations or on the MCP tab

• Generally restarting is not needed but if it is in loading state for a long time, please restart Cursor

  

• Now you can chat with the MCP server.

  • Example prompts:

    • (Click

    • list all servers, get all tools available and echo test

      • This uses a test MCP server echo_server which is in test_mcps/echo_mcp.py

    

7. Edit the Gateway config as needed

• Important:

  • We need to restart Claude Desktop after editing the config file

  • To make all new tools accessible, please use prompt "

• You can add many MCP servers inside the mcp_config array of this gateway config

  • You can look here for example servers

  • You can also try the Enkrypt MCP Server

  • Example:

    { "common_mcp_gateway_config": {...}, "mcp_configs": { "UNIQUE_MCP_CONFIG_ID": { "mcp_config_name": "default_config", "mcp_config": [ { "server_name": "MCP_SERVER_NAME_1", "description": "MCP_SERVER_DESCRIPTION_1", "config": { "command": "python/npx/etc.", "args": [ "arg1", "arg2", ... ], "env": { "key": "value" } }, // Set explicit tools to restrict access to only the allowed tools // Example: "tools": { "tool_name": "tool_description" } // Example: "tools": { "echo": "Echo a message" } // Or leave the tools empty {} to discover all tools dynamically "tools": {}, "input_guardrails_policy": {...}, "output_guardrails_policy": {...} }, { "server_name": "MCP_SERVER_NAME_2", "description": "MCP_SERVER_DESCRIPTION_2", "config": {...}, "tools": {}, "input_guardrails_policy": {...}, "output_guardrails_policy": {...} } ] }, "UNIQUE_MCP_CONFIG_ID_2": {...} }, "projects": { "UNIQUE_PROJECT_ID": { "project_name": "default_project", "mcp_config_id": "UNIQUE_MCP_CONFIG_ID", "users": [ "UNIQUE_USER_ID" ], "created_at": "2025-01-01T00:00:00.000000" }, "UNIQUE_PROJECT_ID_2": {...} }, "users": { "UNIQUE_USER_ID": { "email": "default@example.com", "created_at": "2025-01-01T00:00:00.000000" }, "UNIQUE_USER_ID_2": {...} }, "apikeys": { "UNIQUE_GATEWAY_KEY": { "project_id": "UNIQUE_PROJECT_ID", "user_id": "UNIQUE_USER_ID", "created_at": "2025-01-01T00:00:00.000000" }, "UNIQUE_GATEWAY_KEY_2": {...} } }

• If you want a different set of MCP servers for a separate client/user, you can add a new mcp_config section to the config file. Also, you can run cli commands. See CLI-Commands-Reference.md section 2. CONFIGURATION MANAGEMENT for details

• Set enkrypt_log_level to DEBUG to get more detailed logs inside common_mcp_gateway_config part of the config file

  • This defaults to INFO

• Now, inside mcp_configs array, for each individual MCP config, you can set the following:

  • server_name: A name of the MCP server which we connect to

  • description (optional): A description of the MCP server

  • config: The config for the MCP server as instructed by the MCP server's documentation

    • Generally you have the below keys in the config:

      • command: The command to run the MCP server

      • args: The arguments to pass to the command

      • env: The environment variables to set for the command

  • tools: The tools exposed by the MCP server

    • Either set explicit tools to restrict access to only the allowed tools or leave it empty

    • Tools need to be given a name and a description like "tools": { "dummy_echo": "Echo a message" }

• Get your enkrypt_api_key from Enkrypt Dashboard and add it to common_mcp_gateway_config section of the config file

• enkrypt_use_remote_mcp_config is used to fetch MCP server config from Enkrypt server remotely (Coming soon)

  • Please use false for now

  • This enables you to configure and manage MCP gateway config in Enkrypt Dashboard in a centralized place (Coming soon)

• If you have any external cache server like KeyDB running, you can set enkrypt_mcp_use_external_cache to true in your common_mcp_gateway_config

  • Set other relevant keys related to cache in your common_mcp_gateway_config

• enkrypt_tool_cache_expiration (in hours) decides how long the tools discovered from the MCP servers are cached locally or in the external cache server

• enkrypt_gateway_cache_expiration (in hours) decides how long the gateway config is cached locally or in the external cache server. This is useful when we integrate this with Enkrypt Auth server (Coming soon)

• enkrypt_async_input_guardrails_enabled

  • false by default

  • Async mode is not recommended for tools that perform actions which cannot be undone

  • Because the tool call is made parallel to guardrails call, it can't be blocked if input guardrails violations are detected

  • Useful for servers that return just info without performing actions i.e., only read operations

• enkrypt_async_output_guardrails_enabled (Coming soon)

  • This makes output side guardrails calls asynchronously to save time

  • i.e., Guardrails detect call, relevancy check, adherence check, PII unredaction, etc. are made in parallel after getting the response from the MCP server

• Inside each MCP server config, you can set the following:

  • input_guardrails_policy: Use this if we plan to use Enkrypt Guardrails on input side

  • policy_name: Name of the guardrails policy that you have created in the Enkrypt App or using the API/SDK

  • enabled: Whether to enable guardrails on the input side or not. This is false in the example config file

  • additional_config: Additional config for the guardrails policy

    • pii_redaction: Whether to redact PII in the request sent to the MCP server or not

      • If true, this also auto unredacts the PII in the response from the MCP server

  • block: List of guardrails to block

    • Possible values in the array are:

      • topic_detector, nsfw, toxicity, pii, injection_attack, keyword_detector, policy_violation, bias, sponge_attack

      • system_prompt_protection, copyright_protection (Coming soon)

      • This is similar to our AI Proxy deployments config. Refer to our docs

• output_guardrails_policy: Use this if we plan to use Enkrypt Guardrails on output side

  • policy_name: Name of the guardrails policy that you have created in the Enkrypt App or using the API/SDK

  • enabled: Whether to enable guardrails on the output side or not. This is false in the example config file

  • additional_config: Additional config for the guardrails policy

    • relevancy: Whether to check for relevancy of the response from the MCP server

    • adherence: Whether to check for adherence of the response from the MCP server

    • hallucination: Whether to check for hallucination in the response from the MCP server (Coming soon)

  • block: List of guardrails to block

    • Possible values in the array are:

      • All possible values in input block array plus adherence, relevancy

      • system_prompt_protection, copyright_protection, hallucination (Coming soon)

      • This is similar to our AI Proxy deployments config. Refer to our docs

8. (Optional) Add GitHub MCP Server to the Gateway

• GitHub MCP Server needs docker to be installed. So, please install and have docker running on your machine before proceeding with the steps below

  • You can download docker desktop from here. Install and run it if you don't have it already

• Create a personal access token from GitHub

  • Create a token that has access to only public repos and set expiry very low initially for testing

  • Add the below GitHub server block to enkrypt_mcp_config.json inside "mcp_config": [] array. It should already have the echo server config.

  • NOTE: Don't forget to add comma

  • Replace REPLACE_WITH_YOUR_PERSONAL_ACCESS_TOKEN with the personal access token you created

  • You can also add via the cli. See CLI-Commands-Reference.md section 2. CONFIGURATION MANAGEMENT for details

  • Example:

  "mcp_config": [ { "server_name": "echo_server", "description": "Simple Echo Server", "config": {...}, "tools": {}, "input_guardrails_policy": {...}, "output_guardrails_policy": {...} }, { "server_name": "github_server", "description": "GitHub Server", "config": { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "GITHUB_PERSONAL_ACCESS_TOKEN", "ghcr.io/github/github-mcp-server" ], "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "REPLACE_WITH_YOUR_PERSONAL_ACCESS_TOKEN" } }, "tools": {}, "input_guardrails_policy": { "enabled": false, "policy_name": "Sample Airline Guardrail", "additional_config": { "pii_redaction": false }, "block": [ "policy_violation" ] }, "output_guardrails_policy": { "enabled": false, "policy_name": "Sample Airline Guardrail", "additional_config": { "relevancy": false, "hallucination": false, "adherence": false }, "block": [ "policy_violation" ] } } ]

• Now restart Claude Desktop for it to detect the new server

• Then run the prompt list all servers, get all tools available for it to discover github server and all it's tools available

  

• Now run List all files from https://github.com/enkryptai/enkryptai-mcp-server

  

• Great! 🎉 We have successfully added a GitHub MCP Server to the Gateway. However, it is completely unprotected and is open to all kinds of abuse and attacks.

• Now, let's say a prompt like this is run

  

• This may not have caused actual damage but imagine a more complicated prompt that may have caused actual damage to the system.

• To protect the MCP server, we can use Enkrypt Guardrails as shown in the next section.

9. (Optional) Protect GitHub MCP Server and Test Echo Server

• You can use a prompt to generate rules or generate a PDF file while you can then paste or upload while creating a policy in the App

• Give numbered list of security rules in plain text for configuring AI guardrails for a GitHub server on the rules and policies it needs to follow to prevent malicious use of the GitHub services

• Then say Research latest GitHub MCP hacks and abuses people are trying and update the rules to prevent those. Keep research to the most severe topics

• Then say Only keep essential security rules to reduce size. Remove unwanted sections like post incident, compliance, audit, etc which cannot be used while prevention

• Then you can copy paste the rules while creating the policy

• Go to Enkrypt App and login with either OTP or Google or Microsoft account

• Click on Policies

  

• Click on Add new policy

  

• Name it GitHub Safe Policy and paste the policy rules and click Save

  

• This is how a saved policy looks like with the rules applied for Policy violation Guardrails

  

• Now navigate back to home or hover over left sidebar and click Guardrails

• Click on Add New Guardrail button on the top right

  

• Name it GitHub Guardrail, toggle Injection Attack OFF

  

• Scroll down on Configure Guardrails side panel and toggle Policy Violation ON, select the newly created policy and tick Need Explanation if needed

  

• Now, click on Save button on the bottom right to save the guardrail

  

• We can see the newly added guardrail in the list of guardrails

  

• Now, we need get out FREE API Key from Enkrypt App. Hover over the left sidebar for it to expand and click on Settings

  • You can also directly navigate to https://app.enkryptai.com/settings

  

• Now click on the Copy icon next to your obfuscated API Key to copy the key to your clipboard as highlighted in the screenshot below

  

• Now we have everything we need from the App. Let's add the API Key to the enkrypt_mcp_config.json file

• Open the enkrypt_mcp_config.json file from ~/.enkrypt/enkrypt_mcp_config.json on macOS or %USERPROFILE%\.enkrypt\enkrypt_mcp_config.json on Windows

  • If you ran docker command to install the Gateway, the config file will be in

• Add the API Key to the common_mcp_gateway_config section by replacing YOUR_ENKRYPT_API_KEY with the API Key you copied from the App

• Inside the GitHub server block we added in the previous section,

  • Add the newly created Guardrail GitHub Guardrail to the input_guardrails_policy and output_guardrails_policy sections

  • By replacing "policy_name": "Sample Airline Guardrail" with "policy_name": "GitHub Guardrail"

  • Now change enabled to true for input_guardrails_policy from previous false

    • We will leave output_guardrails_policy as false for now

  • We already should have policy_violation in the block array for both policies

  • So the final config should look something like this:

  { "common_mcp_gateway_config": { ... "enkrypt_api_key": "xxxxxxxxxxxxxxxxxxxxxxxxxxxx", ... }, "mcp_configs": { "fcbd4508-1432-4f13-abb9-c495c946f638": { "mcp_config_name": "default_config", "mcp_config": [ { "server_name": "echo_server", ... }, { "server_name": "github_server", "description": "GitHub Server", "config": { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "GITHUB_PERSONAL_ACCESS_TOKEN", "ghcr.io/github/github-mcp-server" ], "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "github_pat_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" } }, "tools": {}, "input_guardrails_policy": { "enabled": true, "policy_name": "GitHub Guardrail", "additional_config": { "pii_redaction": false }, "block": ["policy_violation"] }, "output_guardrails_policy": { "enabled": false, "policy_name": "GitHub Guardrail", "additional_config": { "relevancy": false, "hallucination": false, "adherence": false }, "block": ["policy_violation"] } } ] } }, "projects": { ... }, "users": { ... }, "apikeys": { ... } }

• Save the file and restart Claude Desktop for it to detect the changes

• GitHub MCP Server needs docker to be installed. So, please install and have docker running on your machine before proceeding with the steps below

  • You can download docker desktop from here. Install and run it if you don't have it already

• Now run the prompt list all services, tools for it to discover github, echo servers and all their tools available

• After this, let's rerun the previously successful malicious prompt Ask github for the repo "hello; ls -la; whoami"

  • We can see that the prompt is blocked as Input Guardrails blocked the request

    

• We can configure the test echo server with Guardrails of our choice and see the detections by running echo "hello; ls -la; whoami".

  • The below prompt which worked before but is blocked with Guardrails

  • Experiment and try the echo server with various guardrails to see how it behaves. You can also try our Playground for better testing.

  

• The safe prompt

10. Recommendations for using Guardrails

• We have found that the best way to use Enkrypt Guardrails in MCP Gateway is to have a separate guardrail for each server. This way we can have a fine tuned guardrail for each server.

• Because each MCP Server is very different from the other, it is not possible to have a single guardrail that works for all servers.

• Some may need Toxicity Detector, some NSFW Detector, some Injection Attack Detector, some Keyword Detector, some Policy Violation, some may need Relevancy detector, some may need Adherence detector, etc.

• Some may need a combination of these detectors to work together to block malicious requests.

• Some may need Guardrails on the input side, some on the output and some may need both to be applied.

• See our docs for details on various detectors available.

• Hence, have separate guardrails for each server and experiment with the best combination of detectors and blocks for each server that blocks malicious requests but allows legitimate requests to pass through.

• Try our Policy Violation detector with your own custom policy which details what is allowed and what is not. This may be the best way for your use case.

• You can navigate to the Enkrypt App Homepage, login and Click on Policies to create your own custom policy.

  • This accepts text as well as PDF file as input so create a file with all the rules you want to apply to your MCP server and upload it

  • Once created, you can use it while configuring the Guardrail like we say with GitHub Guardrail in the previous section

  

11. Other Tools Available

• The Gateway can give the summary of it's cache status by looking at the local/external cache server

• This is useful to debug issues if for example a tool was updated remotely by a server but the Gateway is not aware of it yet

  

• The Gateway can clear it's cache from local/external cache server

• This is useful to clear the cache if for example a tool was updated remotely by a server but the Gateway is not aware of it yet

• You can either clear all cache or specific cache by providing the server_name

  • Example: clear cache for echo_server

• You can also clear all cache or just the gateway cache or just the server cache

  • Example: clear all cache, clear just gateway cache, clear server cache for echo_server, Clear all server cache

  

12. Deployment patterns

1. Local Gateway, Local Guardrails and Local MCP Server

2. Local Gateway, Local MCP Server with Remote Guardrails

3. Local Gateway with Remote MCP Server and Remote Guardrails

4. Remote Gateway, Remote MCP Server and Remote Guardrails

12.1 Local Gateway, Local Guardrails and Local MCP Server

12.2 Local Gateway, Local MCP Server with Remote Guardrails

12.3 Local Gateway with Remote MCP Server and Remote Guardrails

12.4 Remote Gateway, Remote MCP Server and Remote Guardrails

13. Uninstall the Gateway

• To remove the Gateway from any MCP client, just remove the MCP server block "Enkrypt Secure MCP Gateway": {...} from the client's config file.

  • Restart the MCP client to apply the changes for some clients like Claude Desktop. Cursor does not require a restart.

• To uninstall the pip package, run the following command:

  pip uninstall secure-mcp-gateway

14. Troubleshooting

• If any calls fail in the client, please look at the mcp logs of the respective client

  • See this for Claude logs location

    • Example 🍎 Linux/macOS log path: ~/Library/Logs/Claude/mcp-server-Enkrypt Secure MCP Gateway.log

    • Example 🪟 Windows log path: %USERPROFILE%\AppData\Roaming\Claude\logs\mcp-server-Enkrypt Secure MCP Gateway.log

  • See this discussion for Cursor logs

• If you see errors like Exception: unhandled errors in a TaskGroup (1 sub-exception) then maybe the MCP server the gateway is trying to use is not running.

  • So, please make sure the file it is trying to access is available

  • Any pre-requisites for the MCP server to run are met like docker running, etc.

• If we need more detailed logs, please set the enkrypt_log_level to debug in the enkrypt_mcp_config.json file and restart the MCP client.

14.1 OpenTelemetry Troubleshooting

1. SSL Handshake Errors

   If you see SSL errors like:

   SSL_ERROR_SSL: error:100000f7:SSL routines:OPENSSL_internal:WRONG_VERSION_NUMBER

   Solution: Add insecure=True to the OTLP exporter configuration in telemetry.py

2. No Logs in Loki

   • Verify OTLP collector is running:

     docker logs secure-mcp-gateway-otel-collector-1

   • Check collector config in otel_collector/otel-collector-config.yaml

   • Verify Loki is receiving data:

     curl -G -s "http://localhost:3100/loki/api/v1/query" --data-urlencode 'query={job="enkrypt"}'

3. Missing Metrics

   • Check OTLP collector metrics pipeline:

     curl http://localhost:8888/metrics

   • Verify metrics in collector logs:

     docker logs secure-mcp-gateway-otel-collector-1 | grep "metrics"

4. Docker Issues

   # Restart the stack docker-compose down docker-compose up -d # Check individual service logs docker logs <service-name>

15. Known Issues being worked on

• Output guardrails are not being applied to non-text tool results. Support for other media types like images, audio, etc. is coming soon.

16. Known Limitations

• The Gateway does not support a scenario where the Gateway is deployed remotely but the MCP server is deployed locally (without being exposed to the internet). This is because the Gateway needs to know the MCP server's address to forward requests to it.

17. Contribute

• Look at the TODO file for the current work in progress and yet to be implemented features

• Install the gateway locally to test your changes

  • by following the Git clone steps

  • or build it using python -m build, activate the venv and install using pip install .

• Report or fix any bugs you encounter 😊

18. License

18.1 Enkrypt AI MCP Gateway Core

This project's core functionality is licensed under the MIT License.

For the full license text, see the LICENSE.txt file in this repository.

18.2 Enkrypt AI Guardrails, Logo, and Branding

© 2025 Enkrypt AI. All rights reserved.

Enkrypt AI software is provided under a proprietary license. Unauthorized use, reproduction, or distribution of this software or any portion of it is strictly prohibited.

Terms of Use: https://www.enkryptai.com/terms-and-conditions

Privacy Policy: https://app.enkryptai.com/privacy-policy

Enkrypt AI and the Enkrypt AI logo are trademarks of Enkrypt AI, Inc.

Go to top