Is io.github.lacausecrypto/sophon safe to use?

io.github.lacausecrypto/sophon has no known CVEs as of the latest MCPpedia security scan. It does not require authentication, so any local process can connect — keep this in mind in shared environments.

How do I install io.github.lacausecrypto/sophon?

io.github.lacausecrypto/sophon can be installed by cloning its GitHub repository and following the setup instructions in the README.

Is io.github.lacausecrypto/sophon actively maintained?

io.github.lacausecrypto/sophon is actively maintained — last commit was 0 days ago. It has 1 GitHub stars.

io.github.lacausecrypto/sophon

Honest token economics for MCP agents. Rust binary, zero ML, reproducible benchmarks.

ActiveNot tested

Other

★ 1GitHub

28DNo CVEsactive

Quick Install

{
  "mcpServers": {
    "io-github-lacausecrypto-sophon": {
      "command": "<see-readme>",
      "args": []
    }
  }
}

Setup guide

No install config available. Check the server's README for setup instructions.

Are you the author?

Add this badge to your README to show your security score and help users find safe servers.

Embed in your READMEAbout badges →

[![MCPpedia Score](https://mcppedia.org/api/badge/io-github-lacausecrypto-sophon)](https://mcppedia.org/s/io-github-lacausecrypto-sophon)

Should you use this server?

Honest token economics for MCP agents. Rust binary, zero ML, reproducible benchmarks.

✓

Is it safe?

No package registry to scan.

No authentication — any process on your machine can connect.

License not specified.

✓

Is it maintained?

Last commit 0 days ago. 1 stars.

Will it work with my client?

Transport: stdio. Works with Claude Desktop, Cursor, Claude Code, and most MCP clients.

README

Sophon

Honest token economics for MCP agents. One Rust binary. Zero ML. Reproducible benchmarks.

Sophon is a deterministic context layer for agents speaking the Model Context Protocol. It compresses prompts, conversation memory, code digests, file deltas, and shell output — without an embedding model, without an LLM in the compression path, without a GPU, and without API keys. 7.2 MB default Rust binary (25 MB with the optional 11-language tree-sitter AST backend), MCP-native, cl100k_base-accurate.

Every number below links to the reproducible benchmark script that produced it. Every caveat is in BENCHMARK.md — the source of truth for everything Sophon claims.

TL;DR

Sophon v0.3.0 — stable across N=30/60/80 LOCOMO items, zero ML at runtime.

vs competitors (same inputs, same judge, same machine)

System	LOCOMO accuracy	LLM calls	Latency	Binary	Notes
Sophon COMP_LLM	40.0 % (N=80)	~20 / item	~30 s	34 MB	Block-based Haiku summary
Sophon RETR_HASH	32.5 % (N=80)	0	sub-second	34 MB	Deterministic, $0
Sophon RETR_BGE	33.8 % (N=80)	0	sub-second	34 MB	BGE-small ONNX
Sophon COMP_HEUR	23.8 % (N=80)	0	sub-second	7.2 MB	Minimum baseline
mem0-lite	60.0 % (N=15)	~330 / item	8.7 min	Python stack	LLM fact extraction
LLMLingua-2	n/a (compression metric)	0	2 176 ms	280 MB model	Learned compression
FULL (ceiling)	71.2 % (N=80)	0	—	—	Entire conversation

Head-to-heads:

vs LLMLingua-2 on structured prompts: Sophon +8.9 pts saved, 35× faster (§ 7.8.d)
vs mem0-lite on LOCOMO N=15: tie at 60 %, Sophon sub-second vs 8.7 min, 0 vs 330 LLM calls (§ 7.8.e)

LOCOMO accuracy across sample sizes

Stability check: same random.seed(42), same judge, items are nested (N=60 contains N=30's items, etc.).

Condition	N=15	N=30	N=40	N=60	N=80	95 % CI (N=80)
NONE (floor)	20.0 %	—	27.5 %	—	—	—
COMP_HEUR	26.7 %	23.3 %	27.5 %	23.3 %	23.8 %	[15.8 – 34.1 %]
COMP_LLM (v0.3.0)	—	40.0 %	—	38.3 %	40.0 %	[30.0 – 51.0 %]
RETR_HASH	60.0 %	36.7 %	42.5 %	33.3 %	32.5 %	[23.2 – 43.4 %]
RETR_BGE	60.0 %	30.0 %	32.5 %	31.7 %	33.8 %	[24.3 – 44.6 %]
FULL (ceiling)	73.3 %	66.7 %	75.0 %	75.0 %	71.2 %	[60.5 – 80.0 %]

Honest reading:

N=15 numbers were optimistic (too small a sample, easy items). Scaling to N=40 then N=80 reveals the true performance.
COMP_LLM is stable at 40 % across the three scales where it was measured (N=30/60/80). This is the reliable number.
The retriever plateaus around 32–34 % regardless of the embedder (Hash or BGE).

... View full README on GitHub

Test This Server

No automated test available for this server. Check the GitHub README for setup instructions.

Help improve this page

This server is missing a description. Tools and install config are also missing.If you've used it, help the community.

Add information

Score Breakdown

MCPpedia Score

Click each category to see evidence

28D

Last scored just now

How we score →

Security

3/30

No known vulnerabilities.

○

Known CVEs — No package registry to scan

○

Tool safety — No tools to analyze

○

Tool poisoning — No tools to analyze

○

Injection vectors — No tools to analyze

✓

Tool stability — Tool definitions stable

○

Dependency health — No package to analyze

✗

License — No license specified

○

Authentication — No authentication required

○

Repository — active repo

CVEs checked daily via OSV.dev. Score algorithm is open source.

Reviews

Have you used this server?

Share your experience — it helps other developers decide.

How easy was setup?Did it work reliably?How was the documentation?

Frequently Asked Questions

Is io.github.lacausecrypto/sophon safe to use?: io.github.lacausecrypto/sophon has no known CVEs as of the latest MCPpedia security scan. It does not require authentication, so any local process can connect — keep this in mind in shared environments.
How do I install io.github.lacausecrypto/sophon?: io.github.lacausecrypto/sophon can be installed by cloning its GitHub repository and following the setup instructions in the README.
What AI clients work with io.github.lacausecrypto/sophon?: io.github.lacausecrypto/sophon is compatible with Claude Desktop, Cursor, Claude Code, and most MCP clients that support stdio transport. It uses stdio transport.
Is io.github.lacausecrypto/sophon actively maintained?: io.github.lacausecrypto/sophon is actively maintained — last commit was 0 days ago. It has 1 GitHub stars.

Similar servers

View all →

Memory MCP Server

OfficialNo CVEs98A

Persistent memory using a knowledge graph

5 toolsstdio84.0k81.6k/wk2d ago

Context Mode

No CVEs95A

Privacy-first. MCP is the protocol for tool access. We're the virtualization layer for context.

6 toolsremote7.4k10.0k/wk2d ago

Idea Reality MCP Server

No CVEs95A

Pre-build reality check. Scans GitHub, HN, npm, PyPI, Product Hunt — returns 0-100 signal.

1 toolsremote60643.6k/wk3d ago

Hash-verified file editing MCP server with token efficiency hook. 11 tools for AI coding agents.

No CVEs95A

Hash-verified file editing MCP server with token efficiency hook. 11 tools for AI coding agents.

5 toolsremote405698.1k/wk1d ago

MCP Security Weekly

Get CVE alerts and security updates for io.github.lacausecrypto/sophon and similar servers.

Discussion

Start a conversation

Ask a question, share a tip, or report an issue.

Has anyone used this with Cursor?How do you handle auth?Any alternatives?

Sophon

Honest token economics for MCP agents. One Rust binary. Zero ML. Reproducible benchmarks.

Every number below links to the reproducible benchmark script that produced it. Every caveat is in BENCHMARK.md — the source of truth for everything Sophon claims.

TL;DR

Sophon v0.3.0 — stable across N=30/60/80 LOCOMO items, zero ML at runtime.

vs competitors (same inputs, same judge, same machine)

System	LOCOMO accuracy	LLM calls	Latency	Binary	Notes
Sophon COMP_LLM	40.0 % (N=80)	~20 / item	~30 s	34 MB	Block-based Haiku summary
Sophon RETR_HASH	32.5 % (N=80)	0	sub-second	34 MB	Deterministic, $0
Sophon RETR_BGE	33.8 % (N=80)	0	sub-second	34 MB	BGE-small ONNX
Sophon COMP_HEUR	23.8 % (N=80)	0	sub-second	7.2 MB	Minimum baseline
mem0-lite	60.0 % (N=15)	~330 / item	8.7 min	Python stack	LLM fact extraction
LLMLingua-2	n/a (compression metric)	0	2 176 ms	280 MB model	Learned compression
FULL (ceiling)	71.2 % (N=80)	0	—	—	Entire conversation

Head-to-heads:

vs LLMLingua-2 on structured prompts: Sophon +8.9 pts saved, 35× faster (§ 7.8.d)
vs mem0-lite on LOCOMO N=15: tie at 60 %, Sophon sub-second vs 8.7 min, 0 vs 330 LLM calls (§ 7.8.e)

LOCOMO accuracy across sample sizes

Stability check: same random.seed(42), same judge, items are nested (N=60 contains N=30's items, etc.).

Condition	N=15	N=30	N=40	N=60	N=80	95 % CI (N=80)
NONE (floor)	20.0 %	—	27.5 %	—	—	—
COMP_HEUR	26.7 %	23.3 %	27.5 %	23.3 %	23.8 %	[15.8 – 34.1 %]
COMP_LLM (v0.3.0)	—	40.0 %	—	38.3 %	40.0 %	[30.0 – 51.0 %]
RETR_HASH	60.0 %	36.7 %	42.5 %	33.3 %	32.5 %	[23.2 – 43.4 %]
RETR_BGE	60.0 %	30.0 %	32.5 %	31.7 %	33.8 %	[24.3 – 44.6 %]
FULL (ceiling)	73.3 %	66.7 %	75.0 %	75.0 %	71.2 %	[60.5 – 80.0 %]

Honest reading:

N=15 numbers were optimistic (too small a sample, easy items). Scaling to N=40 then N=80 reveals the true performance.
COMP_LLM is stable at 40 % across the three scales where it was measured (N=30/60/80). This is the reliable number.
The retriever plateaus around 32–34 % regardless of the embedder (Hash or BGE).

... View full README on GitHub