TL;DR
The MCP server connects LLM to exterior instruments and information sources by standardized protocols. Public MCP servers present performance reminiscent of internet search, GitHub entry, database queries, and browser automation by structured software definitions.
These servers sometimes run as long-lived stdio processes that reply to software invocation requests. To reliably use them in your purposes or share them throughout your staff, you should deploy them as steady, accessible endpoints.
Make clear Lets you deploy MCP servers as managed endpoints. The platform executes configured MCP processes, handles lifecycle administration, discovers obtainable instruments, and exposes them by APIs.
This tutorial exhibits you the right way to deploy a public MCP server. We use the DuckDuckGo browser server as our reference implementation. The identical method applies to different stdio-based MCP servers, reminiscent of GitHub, Slack, and file system integration.
DuckDuckGo Browser MCP Server
The Duck Duck Go browser mcp server is an open supply MCP implementation that exposes internet search performance as a callable software. This enables the language mannequin to execute search queries by the MCP protocol and acquire structured outcomes.
The server runs as a regular enter/output based mostly course of and supplies instruments reminiscent of: ddg_search To carry out internet searches. Invoking this software returns structured search outcomes that LLMs can use to reply questions or full duties that require present internet info.
This server requires no extra secrets and techniques or exterior configuration, so use it as a reference implementation. The one requirement is to outline an MCP command. config.yamlThis makes it simple to deploy and take a look at on Clarifai.
If you wish to construct a customized MCP server from scratch utilizing your personal instruments and logic, this information Let’s stroll by the method utilizing FastMCP.
Now that we’ve got outlined our reference server, let’s get began..
Arrange the surroundings
Set up the Clarifai Python SDK.
Set the Clarifai Private Entry Token as an surroundings variable. Acquire your PAT out of your Clarifai account’s safety settings.
create a clone runner instance Navigate to your repository and navigate to the MCP server listing in your browser.
The listing comprises the deployment recordsdata.
config.yaml: Set up configuration and MCP server specs
1/mannequin.py: Implementation of mannequin class
necessities.txt: Python dependencies
Configure the deployment
Replace config.yaml along with your Clarifai mannequin identifier and compute settings earlier than importing. This file defines mannequin metadata, MCP server startup instructions, and useful resource necessities. Clarifai makes use of this to begin the MCP server, allocate compute, and expose the server’s instruments by mannequin endpoints.
of mcp_server This part defines the right way to begin the MCP server course of. directions Specify the executable file, argument lists the arguments handed to that executable. On this instance, uvx duckduckgo-mcp-server Begins the DuckDuckGo MCP server as a regular enter/output based mostly course of.
Implementing the mannequin in 1/mannequin.py inherits from StdioMCPModelClass:
StdioMCPModelClass begins the method outlined in config.yamldiscovers obtainable instruments by the MCP protocol and exposes them by deployed mannequin endpoints. No extra implementation is required aside from inheriting from StdioMCPModelClass.
DuckDuckGo MCP Server runs on CPU and requires minimal assets.
Importing and deploying MCP servers
Add your MCP server utilizing the Clarifai CLI.
of –skip_dockerfile A flag is required when importing an MCP server. This command packages the mannequin listing and uploads it to your Clarifai account.
After you add your MCP server, you may deploy it in your compute to run and course of software requests.
go to computing Create a piece and create a brand new cluster. You will see an inventory of accessible situations in several suppliers and areas, together with their {hardware} specs.
Every occasion shows:
Supplier Area Occasion Kind GPU and GPU Reminiscence CPU and System Reminiscence Hourly Pricing
Choose an occasion based mostly on the useful resource necessities outlined in . config.yaml file. For instance, when you specify sure CPU and reminiscence limits, choose situations that meet or exceed these values. Most MCP servers run as light-weight stdio processes, so a GPU is normally not required until the server explicitly depends on a GPU.
After you choose your occasion, configure your node pool. You possibly can set autoscaling parameters reminiscent of minimal and most replicas based mostly in your anticipated workload.
Lastly, create the cluster and node pool and deploy the MCP server on the compute of your selection. Clarifai begins the server utilizing the command outlined within the file. config.yaml Then expose that software by the deployed mannequin’s endpoint.
Observe this information to discover ways to create a devoted computing surroundings and deploy MCP servers to your platform.
Utilizing the deployed MCP server
As soon as the deployment is full, you should use the FastMCP consumer to work together with the MCP server. The consumer connects to the Clarifai endpoint and discovers obtainable instruments.
Exchange the URL along with your deployed MCP server endpoint.
This consumer establishes an HTTP connection to the deployed MCP endpoint and retrieves the software definitions revealed by the DuckDuckGo server. of list_tools() This name verifies that the server is working and that its instruments may be known as.
Integration with LLM
Instruments uncovered by a deployed MCP server can be utilized with any LLM that helps perform calls. Configure MCP purchasers and OpenAI-compatible purchasers to connect with Clarifai MCP endpoints in order that your fashions can uncover and name obtainable instruments.
MCP server has been deployed as an API endpoint. Make clearand its instruments may be accessed and invoked by appropriate LLMs by MCP purchasers.
Often requested questions (FAQ)
Can I deploy any MCP server utilizing this technique?
sure. So long as the MCP server is working as a regular enter/output based mostly course of, the MCP server mcp_server part of config.yaml. After you replace your instructions and arguments and add your mannequin, your server can be uncovered by its personal endpoint.
Is Docker required to deploy MCP Server?
no. When importing an MCP server utilizing the Clarifai CLI, –skip_dockerfile This flag permits deployment with out the necessity for a customized Dockerfile.
Can I exploit the deployed MCP server with any LLM?
sure. LLMs that help perform calls or software calls can use instruments uncovered by the deployed MCP server. The software have to be formatted in keeping with the mannequin’s perform name schema.
Does the MCP server require an API key?
It is dependent upon the server implementation. Some public MCP servers, such because the DuckDuckGo instance used on this information, don’t require extra secrets and techniques. You might also want API credentials outlined in surroundings variables or settings.
lastly
We transformed our stdio-based MCP server right into a publicly accessible API endpoint. Make clear. The software can now be found and known as by any LLM that helps perform calls.
This method lets you transfer your MCP server from native growth to a steady, shareable infrastructure with out altering the core implementation. In case your server is working on stdio, you may bundle, deploy, and publish by Clarifai.
Now you can deploy your personal MCP server, hook up with your fashions, and lengthen your LLM utility with customized instruments and exterior integrations. See beneath for extra examples. Runner instance repository.
