Skip to content

README_EN.md

Latest commit

 

History

History
208 lines (140 loc) · 5.26 KB

README_EN.md

File metadata and controls

208 lines (140 loc) · 5.26 KB
Raw

MySearch Skill

中文说明 · Back to repo

The purpose of skill/ is simple:

  • teach Codex how to install and use MySearch
  • let Claude Code follow a repo link and still install the MCP correctly

This is not the standalone MCP implementation directory.

Think of it as the human and AI readable install-and-use guide:

  • README_EN.md
    • explains how to install
    • explains how to verify
  • SKILL.md
    • explains how to use MySearch after installation

If you are a user, the easiest way to use this

Send this to Codex or Claude Code:

Open skill/README_EN.md and skill/SKILL.md from this repository, install MySearch for me, run install.sh from the repo root if the MCP is not registered yet, then run the health check and smoke tests and tell me the result.

If you only want to share the GitHub link, you can also say:

Please read https://github.com/skernelx/MySearch-Proxy/tree/main/skill and automatically install and verify MySearch for me.

If you already have a deployed MySearch streamableHTTP endpoint, you can also send this:

Register this MySearch MCP URL in Codex and verify it for me: http://127.0.0.1:8000/mcp. If auth is required, use the MYSEARCH_MCP_BEARER_TOKEN environment variable.

If you are an AI assistant, what to do after receiving this directory

1. Treat this as the install-and-use entry, not as a standalone MCP package

If the user gives you any of these:

  • the repository URL
  • the skill/ directory URL
  • a local skill/ path

default to using this folder as the MySearch onboarding entry point.

2. Install the Codex skill first

If the current environment is Codex, run:

bash skill/scripts/install_codex_skill.sh

If the destination already exists and needs replacement:

bash skill/scripts/install_codex_skill.sh --force

After installation, remind the user to:

  • restart Codex

3. Then ensure the MySearch MCP is installed

If the user gave you the source repository, follow the local install path:

From the repository root:

python3 -m venv venv

If the config file is not prepared yet:

cp mysearch/.env.example mysearch/.env

Then fill the environment for the current deployment:

  • MYSEARCH_TAVILY_*
  • MYSEARCH_FIRECRAWL_*
  • optional MYSEARCH_XAI_*

Then run:

./install.sh

Notes:

  • skill/ teaches the assistant how to use MySearch
  • the root install.sh registers the mysearch MCP for Codex / Claude Code
  • both parts matter

If the user gives you a deployed MySearch streamableHTTP URL instead of the source repository, treat it as a remote MCP and do not ask the user to run ./install.sh locally:

codex mcp add mysearch --url http://127.0.0.1:8000/mcp
codex mcp get mysearch

If the endpoint requires bearer auth:

export MYSEARCH_MCP_BEARER_TOKEN=your-token
codex mcp add mysearch \
  --url https://mysearch.example.com/mcp \
  --bearer-token-env-var MYSEARCH_MCP_BEARER_TOKEN
codex mcp get mysearch

Keep the paths separate:

  • local repository install = stdio
  • remote URL attach = streamableHTTP
  • the openclaw/ bundle does not depend on this remote MCP URL

Recommended provider path

The default recommendation is not to hand-fill every official provider key. The recommended setup is:

Why this is better:

  • better for public projects
  • better for team reuse
  • better for AI-driven installation flows

How to verify the installation

Use this order:

codex mcp list
codex mcp get mysearch
python skill/scripts/check_mysearch.py --health-only
python skill/scripts/check_mysearch.py --web-query "OpenAI latest announcements"
python skill/scripts/check_mysearch.py --docs-query "OpenAI Responses API docs"

If X / Social is configured, add:

python skill/scripts/check_mysearch.py --social-query "Model Context Protocol"

If you want to test extraction too:

python skill/scripts/check_mysearch.py \
  --extract-url "https://www.anthropic.com/news/model-context-protocol"

How Claude Code should interpret this skill

This folder currently provides:

  • a local skill installer for Codex
  • shared usage and installation instructions for both Codex and Claude Code

That means:

  • Codex can install the local skill directly
  • Claude Code can still read this README_EN.md and SKILL.md, install the MCP, verify it, and then follow the same usage rules

How the AI should use MySearch after installation

After installation, do not fall back to generic web search by default.

Preferred order:

  1. check mysearch_health
  2. start from search
  3. use extract_url when page content is needed
  4. use research when a lightweight research pack is needed
  5. only fall back to other search tools if MySearch is unavailable or the user explicitly asks for another source

For the full behavior rules, see:

Related docs