mcp-sophon
v0.5.4
Published
Deterministic context compressor for MCP agents. Slots in front of prompt caching, mem0, Letta, or Claude Code — compresses prompts, conversation memory, code digests, file deltas, and shell output. Single Rust binary, zero ML at query time, MCP 2025-06-1
Maintainers
Readme
mcp-sophon
npm wrapper for Sophon.
Honest token economics for MCP agents. One Rust binary. Zero ML. Reproducible benchmarks.
Sophon is a deterministic context layer for MCP-speaking agents: query-driven prompt compression, conversation memory, code navigation, file delta streaming, output compression. Measured 67 % session token savings, ~10 pts more saved than LLMLingua-2 on structured prompts at 35× lower latency, and parity with mem0 on LOCOMO retrieval at sub-second vs 8.7 minutes. Every number reproducible from BENCHMARK.md.
Installing this package downloads a prebuilt native binary from the
matching GitHub release and exposes it as the sophon command.
Install
npm install -g mcp-sophon
sophon --helpSupported platforms (via GitHub Release prebuilds):
| OS | Arch | |---|---| | macOS | arm64, x64 | | Linux | arm64, x64 | | Windows | x64 |
Use as an MCP server
{
"mcpServers": {
"sophon": {
"command": "sophon",
"args": ["serve"]
}
}
}Environment
SOPHON_SKIP_DOWNLOAD=1— do not attempt to download the binary on install (useful in Docker / air-gapped CI). Provide asophonbinary onPATHyourself.SOPHON_REPO=org/repo— override which GitHub repository the postinstall fetches from. Defaults tolacausecrypto/mcp-sophon.
See also
License
MIT
