Introduction

Integrating third-party MCPs into OpenClaw often involves hassles such as long login chains, selecting the wrong regional cluster (China / Europe / US), and inconsistencies between connection configurations and local token caches. The official COROS team has released COROS MCP (coordinates @coros/coros) on SkillHub. It uses the unified gateway https://mcp.coros.com for discovery, then locks login, refresh, MCP calls, and local caches to the CN, EU, or US cluster selected by the gateway. The following sections introduce what it is, what it does, and how to install and use it.

What is This

COROS MCP is an enterprise-sourced Skill maintained by COROS (SkillHub category ai-agent, current version 1.0.1). In one sentence: It installs or refreshes the COROS MCP connection within OpenClaw, automatically binding to the corresponding regional issuer via the global gateway mcp.coros.com, rather than manually writing multiple sets of MCP endpoints.

The SkillHub page is marked with requires_api_key=false, indicating that installing the Skill itself does not require an additional API Key. Actual access to COROS MCP still utilizes OAuth browser login and local token caching (see the script workflow below).

Official Resource Entries:

  • Skill Package Address: https://api.skillhub.cn/coros/coros
  • SkillHub Display Page: https://www.skillhub.cn/skills/coros/coros

Core Functions

Based on the SKILL.md and auxiliary scripts within the Skill, the verified capabilities are as follows.

  1. Gateway Discovery and Region Locking
    Connects to https://mcp.coros.com by default, reads the gateway-selected issuer from discovery, and pins login, token refresh, MCP calls, and cache directories to specific regions. Local states for CN / EU / US are stored in separate directories (~/.coros-mcp-skill-gateway/<region>/) and do not overwrite each other.

  2. Writing OpenClaw MCP Configuration
    Creates or refreshes a MCP entry named coros via apply-openclaw; upon success, you can verify it using openclaw mcp show coros.

  3. Login Methods

    • Single machine or quick refresh: A single apply-openclaw command. It reuses cache if available; otherwise, it outputs a browser login link.
    • Browser might be on another device: Run login-start first, have the user complete login, then run login-finish on the original machine, followed by apply-openclaw.
    • Optional --legacy and --username flags are available for the legacy password flow (password entered interactively).
    • Use login-status to view pending logins and reprint links; logout should only be used when the user explicitly requests clearing the local login state.

  4. MCP Tool Discovery and Invocation
    list-tools (defaults to short-term local cache, --refresh can be added), describe-tool, and call-tool utilize the current stateless /mcp flow; tokens are automatically refreshed when nearing expiration.

  5. Force Region (Optional)
    If a specific cluster must be designated, pass --issuer: https://mcpcn.coros.com, https://mcpeu.coros.com, https://mcpus.coros.com.

Runtime Dependencies (Skill Metadata): python3 and openclaw are required on the local machine.

Installation and Activation

Install Skill via SkillHub

The installation prompt provided on the SkillHub page is: Refer to https://skillhub.cn/install/skillhub.md to install @coros/coros.

  1. Ensure the skillhub CLI is installed (if not installed, execute the official install.sh following the installation documentation).

  2. Install the Skill to the skills directory recognized by the current Agent (must use --dir, otherwise the default ./skills/ might not be loaded). OpenClaw and its variants use their own skills directories; if you are letting an Agent execute this same flow in Cursor, the common path is ~/.cursor/skills/ (depending on your current environment).

skillhub install @coros/coros --dir <your Agent skills directory>

Upon successful installation, the Skill directory contains scripts/coros_mcp_login.py. Subsequent commands should be executed within this Skill directory.

Connect MCP in OpenClaw

Navigate to the installed @coros/coros Skill directory (containing scripts/) and select the workflow as needed.

For cross-device scenarios or when terminal interruption is possible (Two-step login recommended):

python3 scripts/coros_mcp_login.py login-start
python3 scripts/coros_mcp_login.py login-finish
python3 scripts/coros_mcp_login.py apply-openclaw --server-name coros

For single machine or quick refresh (One-step):

python3 scripts/coros_mcp_login.py apply-openclaw --server-name coros

If there is no local token, the script will print a browser login link. Simply complete the COROS login in a browser on your phone or computer.

Verify if the MCP entry has been saved on the OpenClaw side:

openclaw mcp show coros

If Cursor / Codex CLI etc. only have the Skill files installed, they will not automatically replace the aforementioned OpenClaw MCP configuration; this Skill is designed for OpenClaw + COROS MCP Gateway. Other clients need to confirm if they support equivalent MCP login and configuration methods. Differences not documented in the official Skill docs are not expanded upon here.

Typical Usage

After installation and completing apply-openclaw, you can use the script to directly probe the tools currently provided by the server (examples from the official SKILL.md).

List tools (prioritizes local short-term cache):

python3 scripts/coros_mcp_login.py list-tools

When you need to forcefully fetch the latest list from /mcp:

python3 scripts/coros_mcp_login.py list-tools --refresh

View the parameter structure of a specific tool:

python3 scripts/coros_mcp_login.py describe-tool --tool queryUserInfo

Invoke a tool (arguments are a JSON object):

python3 scripts/coros_mcp_login.py call-tool --tool queryUserInfo --arguments-json '{}'

For natural language scenarios, the official recommendation is to first list-tools to match tools, use describe-tool only when parameters are unclear, and then call-tool; only ask the user for missing required fields.

Use Cases and Notes

Who is it for

  • Already using or planning to use COROS MCP in OpenClaw, hoping to use the official gateway for a unified entry point and automatic region selection.
  • Need to maintain token and cache isolation between CN / EU / US, or occasionally force --issuer specification.
  • Want to complete login, refresh, tool listing, and debugging invocation via scripts rather than manually writing MCP configurations.

Limitations and Notes

  • The core link depends on openclaw and the Skill’s built-in helper; do not create a second configuration path. Always use scripts/coros_mcp_login.py under this Skill.
  • The auxiliary script emphasizes not printing tokens to stdout, and error messages are sanitized; however, browser login links should still only be handled by the account holder.
  • SkillHub Statistics (based on captured data): Rating 100000, Downloads 2720; this may differ from the source field shown in community searches. Refer to SkillHub enterprise source information.
  • This Skill solves connection installation and refresh; specific sports data capabilities depend on the current list-tools return from the COROS MCP server. Tool names and fields will evolve with the server; it is advisable to describe-tool or use --refresh before calling.

Conclusion

COROS MCP converges the COROS (Gaoci) MCP integration on the OpenClaw side into: discovering regions via mcp.coros.com, scripted login, writing configuration via apply-openclaw, and providing tool listing and debugging invocation. If you want to stably maintain this connection in your Agent workflow, install @coros/coros from SkillHub and follow the steps in the Skill directory to complete OpenClaw configuration. Official Package Address: https://api.skillhub.cn/coros/coros ; Display Page: https://www.skillhub.cn/skills/coros/coros .