SkillFM Beacon

io.github.ericm1018Published by GitHub account ericm1018v0.1.1Updated Oct 24, 2026

AI health, token usage, LLM cost optimization, BYOK vault, and cleanup audits for MCP agents.

The publisher's own server: RealGeeks Company Registry

Overview

AI-generated overview

A local MCP launcher that reports AI health, token usage and LLM spend, suggests cost savings, and audits agent clutter.

What you can do
Check AI health and session status, and inspect token usage and LLM spend across providers such as OpenAI, Anthropic, DeepSeek and Qwen. Find cost-saving opportunities in provider setup, caching, batching and context cleanup. Monitor provider billing with BYOK keys kept in a local vault. Audit slow-agent clutter such as stale worktrees, oversized logs and bloated context, reporting before any cleanup. Detect supported local agent harnesses and install or remove SkillFM guidance after user approval.
When to use it
Useful when you want to see how many tokens an agent used, why an AI bill is rising, or whether a slow agent is weighed down by bloated context or workspace files. Also relevant if you want to connect provider keys without sending them to a SaaS dashboard.
Requirements
Runs locally over stdio via npx from the npm package @skillfm/mcp, which delegates to @skillfm/local. Activation uses a local sidecar bound to 127.0.0.1 with a discovery file under the user's home directory, and requires the user to approve an OAuth device-flow code in their own browser. Provider keys are optional, for BYOK billing monitoring.
Before you install
The package is a thin public launcher; the platform logic is private, so behavior beyond the documented surface cannot be inspected. Activation requires approving a device code in the browser, and the sidecar runs on the local machine. BYOK provider keys are stored in a local vault, and the README says agents cannot read key values back. Cleanup is described as read-only reporting first, but guidance injection and uninstall modify local agent configuration.

Installation

In SourceWeft

  1. Open SkillFM Beacon 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

SkillFM Beacon MCP

AI health, token usage, cost optimization, BYOK monitoring, and safe cleanup for MCP-capable agents.

SkillFM Beacon gives Claude Code, Cursor, Cline, Claude Desktop, VS Code Copilot, and other MCP clients a practical AI operations layer:

  • Check AI health and session status.
  • Inspect token usage and LLM spend.
  • Find cost-saving opportunities across provider setup, cache usage, batching, and context cleanup.
  • Monitor provider billing with BYOK keys stored in a local vault.
  • Audit slow-agent clutter such as stale worktrees, oversized logs, and bloated context.
  • Discover future SkillFM skills for content, social media, e-commerce, SEO/GEO, and workflow automation as they become available.

This repository is the public distribution shell for SkillFM's MCP marketplace listing. The canonical local runtime is @skillfm/local.

Quick Start

Use SkillFM Beacon as an MCP stdio server:

bash
npx -y @skillfm/mcp@latest

Equivalent explicit form:

bash
npx -y @skillfm/mcp@latest mcp

For MCP clients that accept JSON config:

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

Activation Path

The MCP package delegates to @skillfm/local@latest, so the same package can start the local sidecar and activate SkillFM in the current machine.

Start the local sidecar:

bash
npx -y @skillfm/mcp@latest start

The sidecar binds only to 127.0.0.1 and writes its service discovery file to:

text
~/.skillfm/local.json

Agent-driven activation flow:

  1. Read the sidecar URL from stdout or ~/.skillfm/local.json.
  2. Call POST <sidecar_url>/activate/start.
  3. Show the returned verification URL and real device code to the user.
  4. Poll POST <sidecar_url>/activate/poll until it returns activated: true.
  5. Use GET <sidecar_url>/status, GET <sidecar_url>/skills, or the MCP tools.

Do not fabricate a placeholder code. The device code must come from POST /activate/start.

Useful Commands

bash
# Start MCP stdio servernpx -y @skillfm/mcp@latest
# Start local sidecar for current-chat activationnpx -y @skillfm/mcp@latest start
# Show AI health checkup from the CLInpx -y @skillfm/mcp@latest checkup
# Detect supported local agent harnessesnpx -y @skillfm/mcp@latest safe detect
# Install local SkillFM guidance into supported local agents after user approvalnpx -y @skillfm/mcp@latest safe inject
# Remove injected local guidancenpx -y @skillfm/mcp@latest safe uninstall

What SkillFM Helps With

SkillFM Beacon is for users who ask:

  • "How many tokens did I use today?"
  • "Why is my AI bill going up?"
  • "Can I save money on OpenAI, Anthropic, DeepSeek, Qwen, or other LLM providers?"
  • "Is my agent getting slow because the context or workspace is bloated?"
  • "How do I safely connect a provider key without sending it to a SaaS dashboard?"
  • "What skills can my AI run for content, e-commerce, social media, SEO/GEO, or automation?"

Security Model

  • Local sidecar: The activation sidecar listens on 127.0.0.1, not the public network.
  • BYOK vault: Provider API keys are stored locally by @skillfm/local; agents do not read key values back.
  • No private core in this repo: This repository contains distribution metadata and a thin launcher only. Non-public platform implementation details are not included.
  • User-controlled activation: OAuth device-flow activation requires the user to approve the real device code in their own browser.
  • Read-only cleanup first: Cleanup audits are designed to report slow-agent clutter before any user-directed cleanup action.

See PRIVACY.md and SECURITY.md for the public repository boundary.

Marketplace Keywords

SkillFM Beacon is designed to be discoverable for:

MCP server, AI health checkup, token usage, LLM spend, AI cost optimizer, BYOK, local vault, agent monitoring, slow agent cleanup, context cleanup, Claude Code, Cursor, Cline, Claude Desktop, VS Code Copilot, OpenAI usage, Anthropic usage, DeepSeek usage, Qwen usage.

Public Surface Boundary

This public package is intentionally small. It exists to make SkillFM easy to install from MCP marketplaces and agent directories while keeping core safety and platform logic private.

Public:

  • MCP registry metadata
  • npm launcher package
  • install docs
  • activation flow
  • security model
  • user-facing tool descriptions

Private:

  • non-public platform implementation
  • proprietary evaluation internals
  • private prompts or configuration
  • customer data or logs
  • provider credentials

Marketplace Readiness

  • server.json.name matches package.json.mcpName.
  • The npm package is a thin public launcher for @skillfm/local.
  • publishConfig.access is set to public for scoped npm publishing.
  • Public docs avoid guaranteed income, passive income, and automatic earning claims.
  • The repository contains no backend service code, customer logs, provider keys, or non-public automation logic.

See docs/MARKETPLACE-CHECKLIST.md before submitting to MCP registries or agent directories.

License

MIT

Source: README.md at commit 735a211

Tools

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

Version history

1
  1. v0.1.1LatestSep 16, 2026

Score

50% 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

Other Beacon integrations