FreshContext — Claude Code MCP server: setup and overview
An MCP server that discounts stale documents in retrieval workflows using publication dates. It stops agents from acting on outdated search snippets.
claude mcp add --transport http freshcontext https://api.freshcontext.dev/mcpWhat FreshContext does
Vector retrieval often pulls obsolete documents because cosine similarity only compares text meaning. A blog post from several years ago can easily outrank recent technical documentation. FreshContext inserts an evaluation step between fetching data and agent execution. It modifies the initial score with an exponential decay calculation tailored to the source format, such as blog posts or academic papers.
The core tool in the server is evaluate_context. Agents pass in retrieved candidate texts alongside publication timestamps, source URLs, and task profiles. In return, the server generates an action recommendation, confidence ratings, explicit warnings, and a standard envelope.
Version 0.5.3 marks the final release of freshcontext-mcp, as development on this repository has ended. Existing builds remain under the MIT license, accessible either through a local Node.js process with npx or via the remote endpoint at https://api.freshcontext.dev/mcp.
What it helps with
- Downrank stale search snippets in retrieval pipelines
- Balance semantic similarity against publication age
- Generate structured confidence verdicts for candidate sources
- Flag outdated context before passing it to LLMs
Example prompts for the agent
- Evaluate the freshness of these search results with
evaluate_context. - Check whether our retrieved research links are still current for this citation.
- Rank candidate sources by decay-adjusted relevancy and show any date warnings.
How to install FreshContext
1Platform CLI: remote server
claude mcp add --transport http freshcontext https://api.freshcontext.dev/mcp2Platform CLI
claude mcp add freshcontext -- npx -y freshcontext-mcp1Platform CLI: remote server
codex mcp add freshcontext --url https://api.freshcontext.dev/mcp2Platform CLI
codex mcp add freshcontext -- npx -y freshcontext-mcpConfig file
{
"mcpServers": {
"freshcontext": {
"url": "https://api.freshcontext.dev/mcp"
}
}
}{
"mcpServers": {
"freshcontext": {
"command": "npx",
"args": [
"-y",
"freshcontext-mcp"
]
}
}
}Config file
{
"servers": {
"freshcontext": {
"type": "http",
"url": "https://api.freshcontext.dev/mcp"
}
}
}{
"servers": {
"freshcontext": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"freshcontext-mcp"
]
}
}
}1Platform CLI: remote server
gemini mcp add -t http freshcontext https://api.freshcontext.dev/mcp2Platform CLI
gemini mcp add freshcontext npx -y freshcontext-mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about FreshContext
How to install FreshContext in Claude Code?
Register the remote endpoint by running claude mcp add --transport http freshcontext https://api.freshcontext.dev/mcp. Alternatively, configure a local runtime with claude mcp add freshcontext -- npx -y freshcontext-mcp.
What does FreshContext do?
It scores context candidates on temporal decay rather than just semantic similarity. This prevents language models from generating recommendations based on obsolete web pages.
Is FreshContext still maintained?
Release 0.5.3 is the final version, and active feature work on freshcontext-mcp has stopped. The codebase remains open source under the MIT license and works as published.
