Skip to main content
Glama
ecoclaw-mcp

ecoclaw-mcp

Official
by ecoclaw

ecoclaw-mcp

Your AI setup, owned by you, forever. No subscriptions, no billing, no ongoing relationship required.

This is the core MCP server installed on every EcoFiClaw customer's machine during in-person setup. After that one day, EcoFiClaw is out of the picture. Your skills live on your machine, they get smarter over time, and you own everything.


What this is

EcoFiClaw sets you up with a bundle of AI skills tailored to your business — real estate listing writers, market analysis tools, executive productivity helpers, and more. Those skills are installed locally as simple markdown files in ~/.claude/skills/.

This MCP server is the brain that connects those skills to Claude Desktop. It gives you 6 things you can do with your skills — and crucially, it lets you teach your skills to work the way you work.

Your skills get smarter the more you use them — because whenever something isn't quite right, you can tell Claude to fix it, and it stays fixed permanently. That's the homomorphic part: your software reshapes itself based on your feedback.


Related MCP server: Skills Registry MCP Server

The 6 things you can do

1. See what skills you have

list_skills

Shows everything installed, which version, where it came from, and whether you've customized it. A ✏️ means you have a local modification that overrides the original.

2. Use a skill right now

run_skill({ name: "listing-writer" })

Loads the skill's full instructions into your Claude conversation so Claude knows exactly what to do. Always uses your customized version if you have one.

3. Permanently improve a skill

tune_skill({ name: "listing-writer", feedback: "always end with a P.S. that mentions the neighborhood's walkability" })

This is the powerful one. You tell Claude what's not working, and it rewrites the skill's instructions to fix it — permanently, locally, on your machine. The original version is never touched. Next time you use the skill, your version is used automatically.

Examples of feedback that works:

  • "use a more casual, conversational tone"

  • "never mention specific competitor agencies"

  • "always include a social media caption at the end"

  • "add a section for first-time buyers explaining the process"

4. Create a specialist variant

fork_skill({ name: "listing-writer", fork_name: "listing-writer-luxury" })

Creates an independent copy under a new name. Good when you want listing-writer for standard properties AND listing-writer-luxury for high-end ones — completely separate, each improvable on its own.

5. Share an improvement with the community

contribute({ name: "listing-writer", description: "Added neighborhood walkability P.S. — useful for urban markets" })

If you've improved a skill and think it'd help other agents, this submits a pull request to the upstream repository. Uses the gh CLI installed during setup. Completely optional — you choose what and when to share. The diff is shown before anything is submitted.

6. Keep skills up to date

update_skills

Pulls the latest versions of all installed skills from npm. Skills you've customized are automatically skipped — your tuning is never overwritten. You see exactly what updated and what was preserved.


How customization works (the homomorphic concept)

Every skill is a markdown file (SKILL.md) that tells Claude how to behave — what tone to use, what steps to follow, what to include in outputs. When you tune_skill, Claude reads that file and rewrites it to incorporate your feedback.

The original file in ~/.claude/skills/ is never touched. Your modified version goes into ~/.ecoclaw/skills/. When you use a skill, your version takes priority.

This means:

  • Your customizations are durable — they survive npm updates

  • You can always see what changed (it's just a text file)

  • You can roll back by deleting your local version

  • You can contribute improvements back to the community

It's software that reshapes itself. Your stack, your way.


Your files, your data

Everything lives in two places on your machine:

~/.claude/skills/          # npm-installed skills (managed by npx)
  listing-writer/
    SKILL.md

~/.ecoclaw/                # your stuff (managed by this MCP)
  skills/
    listing-writer/
      SKILL.md             # your customized version (overrides above)
    listing-writer-luxury/
      SKILL.md             # your fork
  config.json              # install record: packages, profile, skill metadata
  history.jsonl            # log of every tune operation

Nothing is sent to EcoFiClaw servers. The only network activity is:

  • tune_skill calls the Anthropic API (same as Claude Desktop itself)

  • contribute submits a GitHub PR if you explicitly ask for it

  • update_skills fetches npm packages if you explicitly ask for it

You own these files. Back them up, put them in git, move them to a new machine — they're yours forever.


Contributing back

If you've improved a skill through tune_skill and think your change would help other agents in your area, run contribute. It:

  1. Shows you the diff between your version and the original

  2. Submits a pull request to the upstream skill repository

  3. The EcoFiClaw community reviews and, if it's good, releases it to everyone

You're not required to contribute anything. But if you do, you make the whole system better for everyone who comes after you.


Getting support

Email: root@ecoficlaw.com

The software is designed to handle most things itself. If something feels off — a skill not doing what you expect, a tune not sticking — try tune_skill again with more specific feedback. Claude is pretty good at understanding what you want.

If you're stuck, reach out. We set this up for you in person and we're happy to help.


Claude Desktop config

Add this to your Claude Desktop settings (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "ecoclaw": {
      "command": "npx",
      "args": ["-y", "@ecoclaw/mcp"],
      "env": {}
    }
  }
}

License

MIT. You own this software. Do what you want with it.

Available Tools

9 tools
build_skill_contextA

Build an instruction string for agent delegation from a list of skill names. Returns a string like "Make sure to use the following skills: 'arxiv', 'fred-economics'" that can be prepended to an agent task.

Inputs: skill_names (required).

contributeA

Share a skill improvement back to the community by submitting a GitHub PR to the upstream repo. Shows you the diff first. Uses the gh CLI (installed during EcoFiClaw setup). Completely optional — you choose when and what to share.

Inputs: name (required), description (required).

discover_skillsA

Scan the current project and installed npm packages for available skills. Returns structured manifests with id, name, description, triggers, dataSources, byokKeys, and author. Looks in /skills/, node_modules/@gonzih/skills-/skills/, and node_modules/@ecoclaw//skills/.

Inputs: dir, filter_by_keys.

fork_skillA

Create an independent copy of a skill under a new name. Use this when you want a specialized variant (e.g. listing-writer-luxury) while keeping the original unchanged. Both versions appear in list_skills.

Inputs: name (required), fork_name.

list_byok_keysA

Scan all discovered skills and report which BYOK API keys are present vs. missing in the current environment. Helps you know what to configure to unlock more skills.

Inputs: no named parameters.

list_skillsA

List all installed skills — both npm-installed (/.claude/skills/) and your local customizations (/.ecoclaw/skills/). Shows name, version, source, and whether you've customized it.

Inputs: no named parameters.

run_skillA

Get the full prompt/instructions for a skill so Claude can use it inline. Checks your user-local version first (~/.ecoclaw/skills/), then falls back to the npm-installed version.

Inputs: name (required), context.

tune_skillA

Apply plain-language feedback to a skill, permanently modifying how it works for you. This is the homomorphic heart — your feedback gets written into the skill's prompt and saved locally. The original npm version is never touched. Examples: 'always use a more casual tone', 'end every output with a P.S.', 'never mention competitors'.

Inputs: name (required), feedback (required).

update_skillsA

Pull the latest npm versions of all installed skills. Skills with local customizations are skipped to protect your tuning — you'll see exactly what was updated and what was preserved.

Inputs: skill.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4/5.0

    Scored across 9 tools

    Disambiguation4/5

    Most tools target distinct actions: listing, discovering, running, tuning, forking, updating, contributing, and building context. The main potential confusion is between list_skills and discover_skills, but their descriptions differentiate installed skills from scanned available skills well.

    Naming Consistency4/5

    Tool names mostly follow a consistent snake_case verb_noun pattern such as list_skills, run_skill, tune_skill, fork_skill, and update_skills. The exception is 'contribute', which breaks the pattern by being a bare verb without a noun.

    Tool Count5/5

    Nine tools is a well-scoped set for a skill-management server. Each tool has a clear role in the discover, configure, customize, update, and share workflow, with no obvious bloat or redundancy.

    Completeness3/5

    The toolkit covers listing, discovering, running, tuning, forking, updating, contributing, and context-building, but lacks obvious install and delete/uninstall operations. Agents cannot fully manage the skill lifecycle when a user asks to remove or newly install a skill, though most customization workflows are supported.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Enables Claude to create, edit, run, and manage reusable skills stored locally, including executing scripts with automatic dependency management and environment variables. Works across all MCP-compatible clients like Cursor, Claude Desktop, and claude.ai.
      5
      31
      MIT
    • -
      license
      Not graded
      quality
      Not graded
      maintenance
      Enables intelligent discovery and management of Claude Skills through semantic search, ratings, favorites, and community curation. Provides tools to search, upload, rate, and organize skills with natural language queries and comprehensive metadata.
      -
    • A
      license
      Not graded
      quality
      D
      maintenance
      Turns Claude-style skills (SKILL.md files with resources) into callable MCP tools for any agent. Discovers skills from a directory, exposes their instructions and resources, and can execute bundled helper scripts.
      224 PyPI
      409
      MIT