Configure Agent MCP & Skill

After enabling MCP in TypeX Desktop, complete two more steps to get your AI assistant working with TypeX:

1. Configure Agent MCP — connect your AI tool to the TypeX MCP endpoint
2. Configure Skill — teach the AI how to use TypeX tools correctly

Both steps are required. MCP provides tool access; the Skill file provides workflows, confirmation rules, and error handling.

Endpoint: http://127.0.0.1:52222/mcp/

---

Configure Agent MCP

TypeX MCP uses Streamable HTTP transport. Point your agent at the local endpoint — no command, args, or env fields are needed.

Before configuring your agent, verify:

1. TypeX desktop is open and MCP is enabled
2. Your agent supports HTTP / Streamable HTTP MCP transport

The examples below are grouped by agent type. Start with general-purpose agents if you use TypeX for everyday messaging and automation. Developer-focused agents are listed separately for coding IDE workflows.

General-purpose agents

Claude Code

Claude Code is a terminal-based AI assistant for everyday tasks — messaging, scheduling, research, and automation.

Config file location:

Scope	Path
Global (all projects)	~/.claude.json → top-level mcpServers
Project (current repo)	.mcp.json in project root

Steps:

1. Create or open the config file for your chosen scope
2. Add the typex entry with "type": "http" (required for Claude Code)
3. Save the file
4. Run claude mcp list in the terminal and confirm typex is listed
5. If missing, run claude /doctor to check for schema errors

{
  "mcpServers": {
    "typex": {
      "type": "http",
      "url": "http://127.0.0.1:52222/mcp/"
    }
  }
}

Alternative — CLI one-liner:

claude mcp add typex --type http --url http://127.0.0.1:52222/mcp/

OpenClaw

OpenClaw is a personal AI agent that runs on your machine and connects to messaging channels, automations, and external tools via MCP.

Config file location:

Scope	Path
Global	~/.openclaw/openclaw.json → mcp.servers

Steps:

1. Open ~/.openclaw/openclaw.json (create it if it does not exist)
2. Add the typex entry under mcp.servers with transport: "streamable-http"
3. Save the file
4. Run openclaw mcp list and confirm typex appears
5. Start a new session (/new in chat) or run openclaw mcp reload to pick up changes

{
  "mcp": {
    "servers": {
      "typex": {
        "url": "http://127.0.0.1:52222/mcp/",
        "transport": "streamable-http"
      }
    }
  }
}

Alternative — CLI one-liner:

openclaw mcp set typex '{"url":"http://127.0.0.1:52222/mcp/","transport":"streamable-http"}'

Hermes

Hermes is a general-purpose AI agent with persistent memory, scheduled tasks, and multi-channel messaging support.

Config file location:

Scope	Path
Global	~/.hermes/config.yaml → mcp_servers

Steps:

1. Open ~/.hermes/config.yaml (run hermes config edit if unsure)
2. Add the typex entry under mcp_servers
3. Save the file
4. In the Hermes chat, run /reload-mcp to apply changes
5. Ask the agent to list MCP tools and confirm TypeX tools are available

mcp_servers:
  typex:
    url: 'http://127.0.0.1:52222/mcp/'

Alternative — CLI one-liner:

hermes mcp add typex --url http://127.0.0.1:52222/mcp/

Developer-focused agents

These agents are built into coding IDEs. Use them if you already operate TypeX from your development environment.

Codex

Codex is OpenAI's coding agent integrated into IDE workflows.

Steps:

1. Open Codex settings and navigate to the MCP server configuration section
2. Add a new HTTP MCP server with URL http://127.0.0.1:52222/mcp/
3. Save and restart Codex if prompted

If your Codex version supports a JSON config file, use:

{
  "mcpServers": {
    "typex": {
      "url": "http://127.0.0.1:52222/mcp/"
    }
  }
}

VS Code (GitHub Copilot Agent)

VS Code with GitHub Copilot Agent is a coding-focused assistant with built-in MCP support.

Config file location:

Scope	Path
Workspace	.vscode/mcp.json
User profile	Run command MCP: Open User Configuration

Steps:

1. Run MCP: Open Workspace Folder MCP Configuration (or the user-scope command above)
2. Add the typex entry under servers (VS Code uses servers, not mcpServers)
3. Save the file
4. Open the Chat panel → check that TypeX MCP tools are available under MCP servers

{
  "servers": {
    "typex": {
      "type": "http",
      "url": "http://127.0.0.1:52222/mcp/"
    }
  }
}

Other MCP-compatible agents

IDE extensions such as Cursor, Trae, and Cline also support MCP. The JSON shape varies slightly by tool:

Agent	Config path	HTTP field
Cursor	~/.cursor/mcp.json or .cursor/mcp.json	url
Trae	.trae/mcp.json	url
Cline	Cline panel → Configure MCP Servers	url + "type": "streamableHttp"

Generic template (works for Cursor, Trae, and similar):

{
  "mcpServers": {
    "typex": {
      "url": "http://127.0.0.1:52222/mcp/"
    }
  }
}

Some agents require extra fields — e.g. Claude Code needs "type": "http", VS Code uses servers instead of mcpServers. Check your agent's MCP documentation and adjust field names accordingly.

---

Configure Skill

The Skill file teaches your AI how to use TypeX MCP tools — search-before-send workflows, confirmation rules, ambiguity handling, and error recovery. Without it, the AI has tool access but may skip important safety checks.

Download the Skill File

• 📥 Download SKILL File (Markdown)

The file contains YAML frontmatter (name: typex-use) and detailed usage instructions. Keep the frontmatter intact when importing.

Import the Skill

General approach (all agents):

1. Download the Skill file (link above)
2. Place it in your agent's skill / rules / instructions directory (see per-agent paths below)
3. Rename to SKILL.md if your agent requires that filename
4. Restart or reload your agent

General-purpose agents

Claude Code

1. Open ~/.claude/CLAUDE.md (global) or CLAUDE.md in your project root
2. Append the full content of the downloaded Skill file under a ## TypeX MCP heading
3. Save the file — Claude Code reads CLAUDE.md on startup

OpenClaw

1. Create the skill directory:

mkdir -p ~/.openclaw/skills/typex-use
cp ~/Downloads/typex-mcp-skill.md ~/.openclaw/skills/typex-use/SKILL.md

2. Optionally enable it in ~/.openclaw/openclaw.json:

{
  "skills": {
    "entries": {
      "typex-use": {
        "enabled": true
      }
    }
  }
}

3. Start a new session (/new) and run openclaw skills list to verify

Hermes

1. Create the skill directory:

mkdir -p ~/.hermes/skills/typex-use
cp ~/Downloads/typex-mcp-skill.md ~/.hermes/skills/typex-use/SKILL.md

2. Alternatively, append the Skill content to ~/.hermes/SOUL.md under a ## TypeX MCP heading
3. Start a new Hermes session to pick up changes

Developer-focused agents

Codex

Import the Skill file as a custom instruction or project-level agent config in Codex settings. Paste the full Markdown content into the instructions field.

VS Code

1. Create or edit .github/copilot-instructions.md in your project root (workspace scope)
2. Paste the full content of the downloaded Skill file
3. Save and start a new Copilot Agent session

Other agents

Agent	Skill import path
Cursor	~/.cursor/skills/typex-use/SKILL.md
Trae	.trae/project_rules.md in project root
Cline / Roo Code	Extension Custom Instructions settings

---

Verify Your Setup

Check MCP connection:

Ask your AI to list available MCP tools. You should see TypeX tools such as message search, send message, and group management.

Check Skill integration:

Try a prompt like:

"Search my TypeX groups for product discussions"

Expected behavior:

• The AI searches before acting (does not guess group names)
• Multiple matches → asks you to choose
• TypeX is not running → reports a clear error and suggests enabling MCP in Settings

If MCP tools appear but behavior is wrong, the Skill is likely missing — complete the Skill configuration above.