groundlink
Groundlink
Documentation
Groundlink MCP Server
Use Groundlink's cited web-search results from any Model Context Protocol (MCP)
host. The server exposes one focused tool, `groundlink_search`, which sends a
query to Groundlink and returns source-bearing results (`title`, `url`,
`snippet`, and `source`) for the model to use when it needs evidence rather
than a guess.
Groundlink currently combines Wikipedia and DuckDuckGo results. Each Groundlink
API key includes 100 free test queries; after that, prepaid usage is
$0.001 per query (one credit per query).
> Package: `groundlink-mcp@0.1.1` is published on npm.
> Source: https://github.com/ohhavefun/groundlink
- Live API documentation: https://9ea69cec60fa01f65bbb647a092bcbb4.ctonew.app/docs
- Pricing and credits: https://9ea69cec60fa01f65bbb647a092bcbb4.ctonew.app/pricing
This repository is the home of Groundlink's official developer integrations.
There are two, both maintained here:
| Integration | What it is | Install |
|---|---|---|
| groundlink-mcp (MCP server) | Exposes Groundlink to any MCP host (Claude, Cursor, …) | `npm install -g groundlink-mcp` (also on the official MCP registry as `io.github.ohhavefun/groundlink`) |
| langchain-groundlink (Python/LangChain) | A `GroundlinkSearchTool` for Python agents | `pip install "langchain-groundlink @ git+https://github.com/ohhavefun/groundlink#subdirectory=langchain-groundlink"` |
Both use the same free trial keys (100 queries) from the Groundlink onboarding,
then prepaid usage at $0.001/query.
Where Groundlink is listed
langchain-groundlink (Python / LangChain)
Use Groundlink's cited search from any Python agent or `BaseTool` pipeline.
The tool returns source-bearing results (`title`, `url`, `snippet`, `source`)
so your agent answers from evidence and can cite it.
Install from GitHub (the package is intentionally not on PyPI):
pip install "langchain-groundlink @ git+https://github.com/ohhavefun/groundlink#subdirectory=langchain-groundlink"Quick usage (10 seconds): get a free key from
the docs, then:
from langchain_groundlink import GroundlinkSearchTool
tool = GroundlinkSearchTool(api_key="glk_your_trial_key") # or export GROUNDLINK_API_KEY
result = tool.invoke({"query": "Who was Ada Lovelace?"})
print(result)Full usage, configuration, and error handling are in
`langchain-groundlink/README.md`.
Groundlink MCP Server
Use Groundlink's cited web-search results from any Model Context Protocol (MCP)
host below.
| Field | Value |
|---|---|
| Tool name | `groundlink_search` |
| Input | `{ query: string, max_results?: number }` |
| `max_results` | 1–10; default 5 (or `GROUNDLINK_MAX_RESULTS`) |
| Output | JSON: `{ query, results: [{ title, url, snippet, source }], meta }` |
Use it for factual questions where the host should return URLs/sources with its
answer. The MCP server is deliberately thin: it forwards the query to the
Groundlink HTTPS API and returns the API response over MCP stdio.
Install and run
You need a Groundlink API key (`glk_...`). Ask the Groundlink operator for one
or obtain one through the Groundlink onboarding flow.
npm install -g groundlink-mcp
export GROUNDLINK_API_KEY=glk_your_key_here
groundlink-mcpAlternatively, an MCP host can execute the package without a global install:
npx -y groundlink-mcpThe process uses stdio; it does not open an HTTP port or print normal output to
stdout. MCP hosts should launch it rather than running it interactively.
Configuration
| Variable | Required | Default | Meaning |
|---|---|---|---|
| `GROUNDLINK_API_KEY` | Yes | — | API key used for every Groundlink request. |
| `GROUNDLINK_BASE_URL` | No | Groundlink live URL | Override only for a compatible deployment. |
| `GROUNDLINK_MAX_RESULTS` | No | `5` | Default results per tool call; capped at `10`. |
Claude Desktop configuration
Add this entry to Claude Desktop's MCP configuration file
and restart Claude Desktop. Keep the API key private: do not commit it to a
repository or share the configuration file.
{
"mcpServers": {
"groundlink": {
"command": "npx",
"args": ["-y", "groundlink-mcp"],
"env": {
"GROUNDLINK_API_KEY": "glk_your_key_here"
}
}
}
}If the package is installed globally, use `"command": "groundlink-mcp"` and
omit `args`. For hosts that need an absolute executable path, point `command`
at the installed `groundlink-mcp` binary.
What happens on errors
- Missing `GROUNDLINK_API_KEY`: the server exits at startup with a clear setup
error.
- API key rejected (401): returned to the MCP host as a tool error; correct the
key rather than retrying.
- No remaining free or prepaid credits (402): returned as a tool error; fund
the key before another search.
- Search-source outage (502): returned as a tool error; a short retry may help.
Marketplace description
Groundlink — cited web search for MCP. Give Claude, Cursor, and other MCP
hosts one `groundlink_search` tool that returns source-bearing results instead
of unsupported factual guesses. Every key starts with 100 free tests, then
usage is $0.001 per query through prepaid credits.
Development and release checks
bun install
bun run typecheck
bun run build # emits executable dist/index.js
bun run test # MCP stdio smoke test
npm run pack:check # inspect the npm tarball without publishingThe npm package intentionally ships only the compiled `dist/` runtime, this
README, and npm's required package metadata. Source and smoke tests stay out of
the tarball.
Frequently asked questions
What is groundlink?
groundlink is Groundlink
How do I install groundlink?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is groundlink open source?
Yes — it is hosted on GitHub at https://github.com/ohhavefun/groundlink.
Related MCP tools
Cognee is the open-source AI memory platform for agents. Give your AI agents persistent long-term memory across sessions with a self-hosted knowledge graph engine.
Automate browser based workflows with AI
Hindsight: Agent Memory That Learns
A privacy-first app that strips AI watermarks from content you own.
Agent framework and applications built upon Qwen>=3.0, featuring Function Calling, MCP, Code Interpreter, RAG, Chrome extension, etc.
The power of Claude Code / GeminiCLI / CodexCLI + [Gemini / OpenAI / OpenRouter / Azure / Grok / Ollama / Custom Model / All Of The Above] working as one.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP