Wolfram Language Paclet Repository
Community-contributed installable additions to the Wolfram Language
Client for Docker MCP Toolkit
Introduction
This notebook has a complete usage example of a thin Model Context Protocol (MCP) client of the Raku paclet "MCPClient".
"MCPClient" provides a set of functions for using MCP server's methods by converting them to objects. Those objects can be used with Wolfram Language's Large Language Model (LLM) framework. (, , .)
MCP default profile in Docker's dashboard
In[1]:=
Install the paclet and load it
In[5]:=
PacletInstall["AntonAntonov/MCPClient"];Needs["AntonAntonov`MCPClient`"];
MCP Setup and initialization
Get a Docker profile file
In[7]:=
profileFile=FileNameJoin[{PacletFind["AntonAntonov/MCPClient"]〚1〛["Location"],"Resources","default_docker_profile.json"}];
Import the profile
In[8]:=
RunProcess[{"/usr/local/bin/docker","mcp","profile","import",profileFile}]
Out[8]=
ExitCode0,StandardOutputImported profile default-docker-profile,StandardError
Set the PATH variable (MacOSX in this example)
In[10]:=
SetEnvironment["PATH""/usr/local/bin:/usr/local/share/bin:/usr/local/sbin:"<>("PATH"/.GetEnvironment[])];"PATH"/.GetEnvironment[]
Out[11]=
/usr/local/bin:/usr/local/share/bin:/usr/local/sbin:/usr/bin:/bin:/usr/sbin:/sbin
Setup MCP server process creation command elements
In[9]:=
cmd={"/usr/local/bin/docker","mcp","gateway","run","--profile","default-docker-profile"};
Create the MCP client object
In[12]:=
client=
[cmd,ProcessEnvironmentGetEnvironment[]];
 | MCPStart |
Function[ |
Out[13]=
Initialize the client
In[14]:=
r=
[client,"Pause"10];
| MCPInitialize |
Function[ |
Out[15]=
Tools discovery and creation
Get the MCP server tools list
In[23]:=
mcpTools=
[client,"Pause"8];Length[mcpTools]
| MCPListTools |
Out[24]=
75
Make a table of the tool records (only using names and descriptions)
In[40]:=
tblTools=mcpTools//Map[KeyTake[#,{"name","description"}]&,#]&//Tabular;Dimensions[tblTools]RandomSample[tblTools,5]//SortBy[#,#name&]&
Out[41]=
{75,2}
Out[42]=
Tabular
 |  |
|
Show tools for official GitHub MCP server
In[43]:=
Select[tblTools,StringContainsQ[#description,"GitHub"]&]
Out[43]=
Tabular
| |
|
Make LLMTool objects
In[44]:=
tools=
[client,mcpTools];Short@tools
| MCPToLLMTool |
Out[45]//Short=
Make a configuration with the tools
In[46]:=
conf=LLMConfiguration["Model"{"OpenAI","gpt-5.3-chat-latest"},"MaxTokens"8192,"Tools"tools];
In[57]:=
conf=LLMConfiguration["Model"{"OpenAI","gpt-5.4-mini"},"MaxTokens"8192,"Tools"tools];
LLM invocations
Find the unaccepted GitHub Pull Requests (PRs) -- using the tool "list_pull_requests"
In[47]:=
LLMSynthesize["Which of my -- antononcube -- GitHub pull requests are pending?",LLMEvaluatorconf]
Out[47]=
You currently have **1 open (pending) pull request** on GitHub:- Repo: ollama/ollama PR #13778 — “Request to add community integration: Raku's ‘WWW::Ollama’” Created: 2026-01-19 Link: https://github.com/ollama/ollama/pull/13778 That’s the only open PR under your username right now. If you expected more, they’re likely already merged, closed, or opened under a different account/org.
Create an LLM function that is assumed to invoke the GitHub MCP server tool "list_branches"
In[101]:=
fGHB=LLMFunction[{"Give the branches of the Repo: `1`.","Separate feature from bugfix branches.",LLMPrompt["NothingElse"]["JSON"]},"JSON",LLMEvaluatorconf];
Get branches data of a repository
Process the result and make a corresponding graph
Stopping the MCP server process
Kill the MCP client process
In[102]:=
MCPKill[client]