naga-config

Naga's own configuration management skill

This skill lets you manage and configure the Naga system itself, including modifying system settings, adding MCP tool services, and importing custom skills.

1. View and modify system configuration

Read current configuration

GET /system/config

Returns the complete system configuration JSON, containing the following main sections:

system — Basic settings (ai_name, voice_enabled, stream_mode, debug, log_level)
api — LLM model settings (api_key, base_url, model, temperature, max_tokens, max_history_rounds)
grag — Knowledge graph settings (enabled, auto_extract, similarity_threshold, neo4j connection)
tts — Speech synthesis settings (default_voice, default_speed, default_language)
online_search — Search engine settings (searxng_url, engines, num_results)
live2d — Live2D avatar settings (enabled, model_path, animation_enabled)
handoff — Tool call loop settings (max_loop_stream, max_loop_non_stream)
voice_realtime — Real-time voice conversation settings (enabled, provider, model, voice)
weather — Weather API settings
mqtt — MQTT IoT settings
crawl4ai — Web crawling settings
computer_control — Computer vision control settings

Modify configuration

POST /system/config
Content-Type: application/json

{完整的配置 JSON 对象}

Important: When modifying configuration, you need to send the complete configuration object. It is recommended to first GET the current configuration, modify the fields you need, and then POST it back.

Common configuration operation examples

Switch LLM model: modify api.model、api.base_url、api.api_key Adjust creativity: modify api.temperature(0.0=precise, 1.0=creative, 2.0=random) Modify AI name: modify system.ai_name Toggle voice: modify system.voice_enabled Toggle knowledge graph: modify grag.enabled and grag.auto_extract Adjust number of search results: modify online_search.num_results

2. View existing MCP tool services

GET /mcp/services

Returns a list of all MCP services, including:

name — Service name
displayName — Display name
description — Description
source — Source (builtin built-in / mcporter external configuration)
available — Whether available

3. Add MCP tool services

MCP services are added via JSON configuration, and the configuration will be written to ~/.mcporter/config.json。

POST /mcp/import
Content-Type: application/json

{
  "name": "服务名称",
  "config": {
    "command": "npx",
    "args": ["-y", "@mcp/server-name"],
    "type": "stdio"
  }
}

MCP configuration format

stdio type(started via Node.js command):

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-xxx"],
  "type": "stdio"
}

SSE/URL type(connected via HTTP):

{
  "url": "https://mcp-server.example.com/sse",
  "type": "sse"
}

Search for available MCP tools

When the user needs a feature but there is currently no corresponding MCP tool, use the built-in web search tool (online_search) to search:

Search keywords:MCP server <功能描述> npm or Model Context Protocol <功能> server
Common MCP services are registered under npm's @modelcontextprotocol/ organization
You can also search on GitHub:modelcontextprotocol site:github.com
Once found, use POST /mcp/import to add the corresponding JSON configuration

4. View skill repository

GET /openclaw/market/items

Returns a list of all installable skills.

Install skills from the skill marketplace

POST /openclaw/market/items/{item_id}/install

5. Import custom skills

POST /skills/import
Content-Type: application/json

{
  "name": "技能名称",
  "content": "技能描述和指令内容（Markdown 格式）"
}

The skill file will be created as skills/{name}/SKILL.md, containing YAML frontmatter metadata and Markdown content. The skill will be automatically loaded into the system prompt on the next conversation.

6. View and modify the system prompt

Get the current system prompt

GET /system/prompt

Set the system prompt

POST /system/prompt
Content-Type: application/json

{
  "content": "新的系统提示词内容"
}

Notes

Web search, URL access, and file access use Naga's built-in clawdbot tool and do not require an additional MCP service
Changes to the configuration take effect immediately, with no restart required
After adding an MCP service, you need to confirm in the service list that its status is available
Sensitive information (such as API keys) should be appropriately masked when replying to users

About this skill