Groq Compound MCP: web search and code via Groq
groq-compound-mcp-server
A Groq MCP server with two tools: a question to the Compound model with web search and a question with Python code execution
- 56 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent's questions and their content go to the external Groq API under your key and spend its quota
- The code execution tool runs Python on Groq's side and the result enters the agent's context
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/groq-compound-mcp-serverDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent's questions and their content go to the external Groq API under your key and spend its quota
- The code execution tool runs Python on Groq's side and the result enters the agent's context
This is third-party code. Review the repository files before installing.
What it does
Groq's server passes the agent's questions to Groq compound models and returns the answer. The ask_with_realtime_information tool fits when the answer depends on fresh facts such as news. The model searches the web itself. The ask_with_code_execution tool suits questions where the model benefits from running Python code, such as intermediate calculations. Both tools take the same parameters. The model parameter picks compound-beta or the quick compound-beta-mini. The mode parameter returns only the answer by default, and verbose mode adds the list of tools the model executed. The include_domains and exclude_domains fields narrow the search by domain. The server runs over stdio with npx groq-compound-mcp-server, and the key goes in the GROQ_API_KEY variable. The repository also holds a Next.js variant for hosting on Vercel at /api/mcp with SSE.
Who it is for. Developers with a Groq key who want an agent to reach fast Groq models with built-in web search and code execution.
Good fit when
- An agent with no web search of its own needs yesterday's news or the latest library version
- A calculation is easier to hand to a model with Python than to do in its head: percentages, statistics, parsing a string
- Search results must come only from certain sites, such as vendor documentation
- A quick answer matters more than depth, and compound-beta-mini is enough
- You want to see which tools the model used in order to check the sources
Not a fit when
- The agent already has web search and a second answer source would only add disagreement
- Data must not go to a third party, since the question goes to the Groq API with the request content
- You need code to run on your machine, while the code executes on Groq's side
- You need stable model names, and the code defines only compound-beta and compound-beta-mini
Example request
Ask Groq what changed in the latest Node.js release, limit the search to nodejs.org and show which tools the model used.MCP
- Transport
- stdio, sse
- Authentication
- API key
Environment variables
GROQ_API_KEYrequired, secret- Groq API key
REDIS_URLsecret- Redis address for SSE when hosted on Vercel
Limitations
Node.js 18 or newer and a Groq key in GROQ_API_KEY are required. A question goes to the Groq API together with its content and spends the quota of your key. The README lists only two tools and does not describe model limits. Hosting on Vercel recommends Redis through REDIS_URL. The last commit was made in August 2026, and the model names are set in the server code. SSE mode on Vercel is served at `/api/mcp/sse` by the deployed Next.js app.
How to disable. Remove the groq-compound entry from the MCP client configuration. A global package comes off with npm uninstall -g groq-compound-mcp-server, and the GROQ_API_KEY is revoked in the Groq console.
FAQ
How do I connect Groq Compound MCP to Claude Code?
Add an MCP server with the npx command and the arguments -y and groq-compound-mcp-server, and pass GROQ_API_KEY in env. The README gives a JSON fragment for the client settings file and no ready claude mcp add command.
How do the two tools differ?
ask_with_realtime_information answers with fresh facts from web search, while ask_with_code_execution helps the model solve a task by running Python code. Their parameters are the same.
What does verbose mode do?
The answer gains a list of the tools the model executed. It is long, so enable it only when needed and try the default minimal mode first.
Can I host the server on Vercel?
Yes, the repository builds as a Next.js app with build:vercel and the server answers at /api/mcp. GROQ_API_KEY and preferably REDIS_URL go in the Vercel project settings.
Where is the Groq key set?
The key is set with the GROQ_API_KEY variable in the env block of the MCP client configuration. The server creates the Groq client lazily, on the first question.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server searches the web with Grok, retains sources and extracts page content through available retrieval providers.
A deep research MCP server: multi-source search, result fusion and cited synthesis through a Qwen3 model on OpenRouter
A web search MCP server that gives the agent full page text at once: Stack Overflow answers, GitHub threads, arXiv papers
Metasearch for agents: one MCP server and HTTP API merge results from many search providers into unified JSON with deduplication.
A unified search MCP server: one interface to Tavily, Brave, Kagi, Exa, Linkup, Firecrawl and GitHub search
Linkup MCP server: real-time web search, page fetching and long research tasks with sources for an agent
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Groq Compound MCP: web search and code via Groq (this card) | yes | API key | 56 | |
| Perplexity MCP Server | yes | API key | 2,624 | search and reasoning through the Perplexity API rather than Groq compound models with Python |
| Tavily MCP Server | yes | API key | 2,489 | returns search results and page text while the agent's own model forms the answer |
README in short
The README briefly names the two tools, requirements and a configuration for an MCP client. It then covers hosting on Vercel with environment variables and local development through the Vercel CLI.