MCP Server Integration

SkillForge exposes a Model Context Protocol (MCP) server so any MCP-capable coding agent can invoke all package management operations directly: installing skills mid-session, listing what's available, validating skill quality, and generating suggestions.

The server is agent-agnostic. It communicates over stdio and speaks plain MCP, so Claude Code, Codex, Cursor, Windsurf, Zed, and any other MCP client all work the same way. All tools except skillforge_suggest are pure package-management operations (filesystem + registry) and call no LLM; their behavior is identical regardless of which agent invokes them. Only skillforge_suggest reaches an LLM, and only it needs an API key.

Setup

1. Install SkillForge

npm install -g skillforge

Or from source:

git clone https://github.com/danbenba/skillforge.git
cd skillforge
npm install && npm run build
npm link

2. Register the server with your agent

The server is launched with skillforge mcp (a stdio server). Point your agent's MCP config at that command.

Claude Code: .mcp.json in the project root, or the global ~/.claude/mcp.json:

{
  "mcpServers": {
    "skillforge": {
      "command": "skillforge",
      "args": ["mcp"]
    }
  }
}

Codex: ~/.codex/config.toml:

[mcp_servers.skillforge]
command = "skillforge"
args = ["mcp"]

The ANTHROPIC_API_KEY environment variable is only needed if you use the skillforge_suggest tool. Add it via the config's env block:

{
  "mcpServers": {
    "skillforge": {
      "command": "skillforge",
      "args": ["mcp"],
      "env": {
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}"
      }
    }
  }
}

# Codex equivalent
[mcp_servers.skillforge]
command = "skillforge"
args = ["mcp"]
env = { ANTHROPIC_API_KEY = "sk-ant-..." }

To point skillforge_suggest at a custom Anthropic-compatible endpoint instead, set ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and optionally SKILLPM_SUGGEST_MODEL in the same env block. See suggest.md.

3. Restart your agent

After saving the config, restart the agent (or reload its MCP servers). The SkillForge tools will appear in the tool list.

Transport

The MCP server uses StdioServerTransport: it communicates over stdin/stdout, the standard for local MCP servers. There is no HTTP server, no port, and no authentication needed.

Available Tools

`skillforge_validate`

Validate a skill folder and return its format conformance score.

Input:

Parameter	Type	Required	Description
`path`	`string`	Yes	Absolute or relative path to the skill folder

Output:

{
  "skill": "forensics-agent",
  "score": 91,
  "diagnostics": [
    { "severity": "pass", "message": "YAML frontmatter valid", "check": "yaml-frontmatter" },
    { "severity": "pass", "message": "name field present", "check": "name-present" }
  ],
  "specVersion": "1.0",
  "passCount": 7,
  "warnCount": 0,
  "errorCount": 0
}

Example prompt to Claude Code:

"Validate the skill at ./forensics-agent and tell me if it's ready to install."

`skillforge_install`

Install a skill from a local directory or a git+https:// URL.

Input:

Parameter	Type	Required	Default	Description
`source`	`string`	Yes	(none)	Local path or `git+https://` URL
`scope`	`"global" \| "shared" \| "project"`	No	`"project"`	Installation scope
`force`	`boolean`	No	`false`	Overwrite if already installed

Output:

{
  "installed": true,
  "skillName": "forensics-agent",
  "scope": "project",
  "score": 91,
  "diagnostics": []
}

Example prompt to Claude Code:

"Install the forensics-agent skill from git+https://github.com/acme/forensics-agent at project scope."

`skillforge_uninstall`

Remove an installed skill from a scope.

Input:

Parameter	Type	Required	Default	Description
`skillName`	`string`	Yes	(none)	Name of the skill to remove
`scope`	`"global" \| "shared" \| "project"`	No	`"project"`	Scope to remove from

Output:

{
  "removed": true,
  "skillName": "forensics-agent",
  "scope": "project"
}

`skillforge_list`

List installed skills across scopes.

Input:

Parameter	Type	Required	Description
`scope`	`"global" \| "shared" \| "project"`	No	Filter to a specific scope. Omit to show all.

Output:

[
  {
    "level": "global",
    "skills": []
  },
  {
    "level": "shared",
    "skills": []
  },
  {
    "level": "project",
    "skills": [
      {
        "name": "forensics-agent",
        "version": "1.0.0",
        "source": "community",
        "sourceUrl": "git+https://github.com/acme/forensics-agent",
        "installedAt": "2026-03-26T00:00:00.000Z",
        "specVersion": "1.0",
        "score": 91,
        "path": "skills/forensics-agent"
      }
    ]
  }
]

Example prompt to Claude Code:

"What skills do I have installed for this project?"

`skillforge_suggest`

Generate AI-powered skill suggestions based on project context.

Input:

Parameter	Type	Required	Description
`projectPath`	`string`	No	Path to project. Defaults to the server's working directory.

Output:

{
  "proposals": [
    {
      "skillName": "forensics-agent",
      "reason": "Your README mentions log analysis and debugging production incidents",
      "suggestedScope": "project",
      "available": true
    },
    {
      "skillName": "test-writer",
      "reason": "package.json has extensive test scripts",
      "suggestedScope": "project",
      "available": true
    }
  ]
}

Requirements: ANTHROPIC_API_KEY must be set in the server's environment (see setup above). Alternatively, point it at a custom Anthropic-compatible endpoint with ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN; see suggest.md.

Example prompt to Claude Code:

"Suggest skills I should install for this project based on the codebase."

`skillforge_search`

Search the SkillForge registry for skills.

Status: Stub; returns an empty result set. Full registry search is not yet implemented.

Input:

Parameter	Type	Required	Description
`query`	`string`	Yes	Search query

Output:

{
  "results": [],
  "total": 0,
  "message": "Registry search is not yet available in this version."
}

Available Resources

For MCP clients that support resource reading, the server exposes:

URI	Content
`skillforge://docs/skills-sh-api`	Mandatory reference for the skills.sh API — bundled snapshot of skills.sh/docs/api (Vercel OIDC authentication, endpoints, rate limits, response shapes). Read it before any request to skills.sh. Served with `priority: 1` annotations (effectively required).
`skill://index.json`	SEP-2640 discovery document: the most installed skills of the registry
`skill://<name>/SKILL.md`	The SKILL.md of a registry skill
`skill://<name>/<path>`	Any auxiliary file of a registry skill

Working Directory

The MCP server resolves the project scope relative to its working directory when it starts. When launched by Claude Code, this is typically the project root.

If you're using the server for a different project than where it started, pass an explicit projectPath to skillforge_suggest, or use absolute paths with skillforge_install.

Example Claude Code Session

User: What skills do I have installed?

Claude: [calls skillforge_list]
You have 1 skill installed at project scope:
- forensics-agent (score: 91/100, community)

User: Install the test-writer skill from my ~/skills directory.

Claude: [calls skillforge_install with source=~/skills/test-writer, scope=project]
Installed "test-writer" at project scope. Score: 84/100.

User: Suggest some more skills for this project.

Claude: [calls skillforge_suggest]
Based on your project context, I'd suggest:
1. code-reviewer: your package.json shows a TypeScript project with a review workflow
2. docs-generator: your README mentions documentation as a goal

Would you like me to install either of these?

Troubleshooting

"Could not connect to MCP server"

Verify the path in args is absolute and correct
Run node /path/to/SkillForge/dist/mcp/server.js directly and check for errors
Make sure npm run build was run after any changes

"ANTHROPIC_API_KEY is required"

The skillforge_suggest tool requires the API key in the server's environment
Add it to the env section of your .mcp.json

Tools not appearing in Claude Code

Restart Claude Code after modifying .mcp.json
Check that the JSON is valid (no trailing commas, etc.)
Check Claude Code's MCP server logs

Remote mode (claude.ai custom connector)

The same server also runs over Streamable HTTP for claude.ai:

cp .env.example .env   # set SKILLFORGE_ENV, SKILLFORGE_PORT, SKILLFORGE_PUBLIC_URL
skillforge serve       # or: docker compose up -d --build

Then on claude.ai: Settings > Connectors > Add custom connector with the URL https://<your-domain>/mcp (no authentication, unless SKILLFORGE_AUTH_TOKEN is set).

Remote mode exposes the catalog tools only (skillforge_start, skillforge_search, skillforge_skillset_search, skillforge_inspect, skillforge_activate, skillforge_file, skillforge_compare, skillforge_validate_remote); the filesystem install tools are stdio-only, because claude.ai has no access to your disk. On claude.ai, installation is virtual: skillforge_activate returns the complete SKILL.md and Claude follows it verbatim for the rest of the conversation. For a permanent install, Claude recreates the skill in the claude.ai skill panel and you click "Copy to your skills".

Navigation Menu

Repository navigation

MCP Server Integration

Setup

1. Install SkillForge

2. Register the server with your agent

3. Restart your agent

Transport

Available Tools

`skillforge_validate`

`skillforge_install`

`skillforge_uninstall`

`skillforge_list`

`skillforge_suggest`

`skillforge_search`

Available Resources

Working Directory

Example Claude Code Session

Troubleshooting

Remote mode (claude.ai custom connector)

FilesExpand file tree

mcp.md

Latest commit

History

mcp.md

File metadata and controls

MCP Server Integration

Setup

1. Install SkillForge

2. Register the server with your agent

3. Restart your agent

Transport

Available Tools

skillforge_validate

skillforge_install

skillforge_uninstall

skillforge_list

skillforge_suggest

skillforge_search

Available Resources

Working Directory

Example Claude Code Session

Troubleshooting

Remote mode (claude.ai custom connector)

`skillforge_validate`

`skillforge_install`

`skillforge_uninstall`

`skillforge_list`

`skillforge_suggest`

`skillforge_search`