npx -y @vexlo/mcp to run the Vexlo MCP server. Your MCP client launches it as a stdio
process, which reads recorded execution from your Vexlo server over HTTP. This guide uses
Claude Code; other MCP clients use the same command, server URL and read credential.
Prerequisites
- A Vexlo server with recorded data and a read-role token for the relevant teams.
- Node.js 20 or later and npm.
- Claude Code installed and authenticated for the walkthrough.
Configure Claude Code
From your application project, set the server address and read token, replacing the example values:http://127.0.0.1:8080, without /v1. Both variables are
required in every shell that starts Claude Code; the project configuration resolves them when
launching the MCP process.
Register Vexlo for this project:
.mcp.json. The quoted variable references remain
in that file and expand from the environment when the server launches, so the command above
does not embed the token’s value in the project configuration.
Start Claude Code from the project:
--strict-mcp-config excludes MCP configuration
from other sources.
Approve a project-scoped server when Claude Code requests trust, then use /mcp to inspect its
status. Other clients have their own stdio registration settings; .mcp.json is specific to
this Claude Code setup.
Verify the connection
Ask Claude Code to callvexlo_check_coverage with an empty argument object. An example
response:
Connection failures
Startup validates configuration, checks/v1/capabilities, then checks the token’s identity
through /v1/whoami before registering tools.
A startup with empty required variables returned:
Develop the MCP server
To modify or test the MCP server, build a local checkout with Git:dist/index.js. Return to your application project in the same
terminal, keeping VEXLO_MCP_ENTRY, VEXLO_API_URL and VEXLO_TOKEN set. If the project
already has a vexlo entry, remove it first:
