Agentskin

io.github.Shawn5centsPublished by GitHub account Shawn5centsv4.0.1Updated Oct 24, 2026

The Semantic Layer for AI Agents. Prunes noisy API data by 70%+ for token-efficient reasoning.

VerifiedSTDIODesktop onlyWeb Search & ScrapingDeveloper Tools

Overview

AI-generated overview

Local Node.js MCP server that compresses JSON, CLI output, plain text and fetched web pages before they enter an AI assistant's context.

What you can do
  • Compress JSON, CLI output or plain text through one auto-detecting entry point
  • Fetch a public URL and apply URL-specific pruning rules to the response
  • Reduce terminal output with rule-driven reducers
  • Run diagnostics such as classifying a URL, stripping ANSI codes, estimating token counts and inspecting reasoning
When to use it
Useful when an assistant works with large API responses, verbose JSON, long terminal logs or fetched web pages and you want to cut token usage while keeping the information the task needs.
Requirements
Runs locally over stdio via npx, so Node.js is needed. Optional search and media API keys can be supplied through environment variables: APILAYER_KEY, BRAVE_SEARCH_API_KEY, SERPER_API_KEY and TAVILY_API_KEY. Network access is required for URL fetching.
Before you install
It fetches public URLs on the assistant's behalf; the README mentions SSRF checks, response-size limits, rate limiting and timeouts, but fetched content still leaves your machine. Optional API keys are passed as environment variables and may incur provider usage costs. Compression is lossy by design, so verify that needed details survive.

Installation

In SourceWeft

  1. Open Agentskin in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

AgentSkin

[npm version] [License: MIT]

Context middleware for AI agents. AgentSkin removes low-value API, JSON, and terminal noise before it enters model context.

The frozen AgentSkin 5.1 offline core evaluation also has an accepted Sparse Env execution path. AgentSkin preserves the workload correctness contract while Sparse Env owns cold/warm environment selection and lifecycle; see agentskin/docs/SPARSE_ENV.md.

What it does

AgentSkin has three primary workflows:

  • compress — one front door that auto-detects JSON, CLI output, or plain text.
  • fetch_optimized_data — fetches a public URL and applies URL-specific pruning rules when available.
  • reduce — compacts terminal output using Tokenjuice's rule-driven reducers.

Five advanced tools remain for compatibility and diagnostics: apply_json_semantic, classify_url, strip_ansi, estimate_tokens, and skin_reasoning.

Quick start

bash
npx -y agentskin@latest

MCP config:

json
{  "mcpServers": {    "agentskin": {      "command": "npx",      "args": ["-y", "agentskin@latest"]    }  }}

Design rule

Fidelity first, compression second. Explicit signals and URL rules are authoritative. Generic keys such as id, name, and url are only fallback signals when no explicit rule exists.

A reduction is useful only when the information required by the task survives it.

Current verification

The release gate runs the AgentSkin-owned test suite, including MCP startup/security and live HTTP integration tests. Run it with:

bash
cd agentskinnpm cinpm test

Benchmarks are workload-specific. The test suite prints measured reductions for its fixtures instead of claiming one universal savings percentage.

Runtime

  • Node.js MCP server over stdio
  • Tokenjuice is a normal npm dependency for CLI reduction
  • URL-specific semantic pruning rules for structured APIs
  • SSRF checks for fetches, response-size limits, rate limiting, Zod validation, and processing timeouts

Development

The npm package lives in agentskin/ inside this repository.

bash
cd agentskinnpm cinpm testnpm run lintnpm pack --dry-run

Credits

AgentSkin semantic pruning and MCP integration: Shawn Nichols Sr. / Nichols Transco LLC.

CLI reduction uses Tokenjuice by Vincent Koc under its MIT license.

Source: README.md at commit 0c35b05

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v4.0.1LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.