Skip to content
MCP server

Semantic Cache MCP

File cache for coding agents that returns tiny re-reads and diffs

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, macOS, Linux
Runs
Locally
Version
0.7.0
Listing updated
today

Similar MCP servers

About Semantic Cache MCP

Semantic Cache MCP is an open-source MCP server for developer tools. It runs locally on your machine and is started by your AI client.

Coding agents re-read the same files before and after every edit, and a normal read resends the whole file each time. This server gives each file a content hash on first read; when the agent passes the hash back, an unchanged file comes back as a short notice, and an edited file comes back as a diff of just the changed lines. The README measures this on a 7,891-token file: 77 tokens for an unchanged re-read and 137 after a one-line edit.

The hash is checked against the bytes on disk on every read, so stale content is not served, and an agent that lost its context after compaction can simply read again without the hash. Files larger than the size limit return an outline of their definitions instead of a blind cut. Search is BM25 keyword ranking, not embeddings, so there is no model download, GPU or API key.

Everything stays local in a SQLite file, served by one background daemon shared by all clients. The only network request is a one-time tokenizer file download, checked against a SHA-256 hash.

Install

Run in a terminal

claude mcp add semantic-cache-mcp -- uvx semantic-cache-mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "semantic-cache-mcp": {
      "command": "uvx",
      "args": [
        "semantic-cache-mcp"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "semantic-cache-mcp": {
      "command": "uvx",
      "args": [
        "semantic-cache-mcp"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "semantic-cache-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "semantic-cache-mcp"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "semantic-cache-mcp": {
      "command": "uvx",
      "args": [
        "semantic-cache-mcp"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "semantic-cache-mcp": {
      "command": "uvx",
      "args": [
        "semantic-cache-mcp"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Tools this server provides (14)

Taken from the project documentation; the running server may differ.

read
Reads a file, returning the full text, a short unchanged notice, or a diff depending on the hash passed back.
batch_read
Reads several files under a token budget.
read_image
Reads an image file.
write
Writes a file.
edit
Edits a file.
batch_edit
Applies several edits.
edit_preview
Previews an edit as a dry run.
delete
Deletes a file.
warm
Loads files into the cache ahead of time.
grep
Exact search over the cache.
search
BM25 keyword search over the cache.
glob
Matches file paths by pattern.
stats
Reports cache statistics.
clear
Clears the cache.

What it does

Hash-based re-reads
Unchanged files return a short notice and edited files return a diff against the hash the agent holds.
Cache-wide search
grep, BM25 keyword search and glob run over cached files.
Atomic edits
Edit tools include a dry-run preview and batch edits.
Outline for big files
Files over the size limit return a map of definitions.
Local-only storage
A SQLite cache and one shared daemon; no code is sent anywhere.

Good for

  • Cut token spend when an agent keeps re-opening the same files
  • Keep long coding sessions affordable across context compactions
  • Preview an edit before writing it with edit_preview
  • Batch-read several files within a token budget

Pricing

Open source

MIT licence

Semantic Cache MCP FAQ

Is the search semantic?

No. search is BM25 keyword ranking and grep is exact, so no embedding model, GPU or API key is needed.

Does it send my code anywhere?

No. The only request is a one-time download of the o200k tokenizer file, verified by SHA-256.

How do I install it?

With uv: claude mcp add semantic-cache --scope user -- uvx semantic-cache-mcp, or an mcpServers entry that runs uvx semantic-cache-mcp.

What licence is it under?

MIT.

Is Semantic Cache MCP free?

Yes. Semantic Cache MCP is open source, so you can use and self-host it for free.

How do I add Semantic Cache MCP to Claude Code?

Run this command in your terminal: claude mcp add semantic-cache-mcp -- uvx semantic-cache-mcp

What are the best alternatives to Semantic Cache MCP?

The closest alternatives to Semantic Cache MCP are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers