Currents MCP Server

Give your AI coding agents full visibility into your CI test results. The Currents MCP Server connects tools like Cursor and Claude directly to your Currents dashboard, so agents can diagnose flaky tests, pinpoint failures, and act on real execution data -- without leaving your editor.

• Query runs, spec files, and individual test results from CI
• Surface error trends and performance metrics across your test suite
• Manage quarantine rules, webhooks, and project settings programmatically
• Let agents fix what's broken using actual test output, not guesswork

Tools

Skills

Agent skills that teach AI agents multi-step Currents workflows.

The server publishes every skill as an MCP resource, so a connected agent can

list and read them without a local copy:

skill://currents/collect-evidence/SKILL.md
skill://currents/collect-evidence/references/instrumentation.md

Agents that load skills only from disk still need a copy — put the skill

directory in the agent's skills location (e.g. `.claude/skills/` for Claude

Code).

Setup

API Key

Get a Currents API key by following the instructions here.

Usage with Cursor

1. Go to Cursor Settings > MCP > Enable

2. Add the following to your `mcp.json`.

{
  "mcpServers": {
    "currents": {
      "command": "npx",
      "args": ["-y", "@currents/mcp"],
      "env": {
        "CURRENTS_API_KEY": "your-api-key"
      }
    }
  }
}

Claude

Run this command to add Currents MCP to Claude Code

claude mcp add --transport stdio currents --env CURRENTS_API_KEY= -- npx -y @currents/mcp

Add the following to enable Currents MCP on Claude Desktop (edit `claude_desktop_config.json` file):

{
  "mcpServers": {
    "currents": {
      "command": "npx",
      "args": ["-y", "@currents/mcp"],
      "env": {
        "CURRENTS_API_KEY": "your-api-key"
      }
    }
  }
}

Remote (hosted) MCP endpoint

In addition to the local stdio transport above, the same server can run as a hosted

Streamable HTTP endpoint (e.g. `https://mcp.currents.dev/mcp`) for use with remote

connectors such as the Claude web/mobile apps.

The hosted server performs no authentication of its own. Each request must carry

your Currents API key as a Bearer token, which is passed through to the Currents API:

Authorization: Bearer

Example client config (remote connector):

{
  "mcpServers": {
    "currents": {
      "url": "https://mcp.currents.dev/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

Running the HTTP server yourself:

# from the mcp-server package
npm run build && PORT=3000 npm run start:http
# or via Docker (serves /mcp, exposes the configured PORT)
docker build -t currents-mcp . && docker run -p 3000:3000 currents-mcp

The Node server speaks plain HTTP; TLS and the public domain terminate at the

reverse proxy / load balancer in front of the container. A `GET /healthz` endpoint

is provided for liveness checks.

⚠️ Notice

By connecting AI tools (e.g., via MCP) to Currents, you are granting them access to your API key, test results and CI metadata. It is your responsibility to vet any AI agents or services you use, and to ensure they handle your data securely.

References

• Currents
• Currents Documentation
• Contribution Guide
• Releasing and Publishing
• License