evaficy-mcp

An MCP (Model Context Protocol) server for Evaficy Smart Test. It lets an AI client (Claude Code, Cline, Cursor, etc.) read your projects, test cases and run results directly from your editor.

This is v0.5.1: six read-only tools plus one write tool, generate_test_cases.

Tools

| Tool | Type | What it does | |---|---|---| | list_projects | read | List all projects you own or belong to | | get_project | read | Full detail for one project (members, environments, client) | | list_test_cases | read | List test-case scenarios for a project | | get_test_case | read | Full scenario detail: steps, test data, validation status | | list_runs | read | List test runs, optionally filtered by project/status | | get_run_results | read | A run's execution detail plus its condensed report | | generate_test_cases | write | Generate ~7 AI test cases from a requirement, optionally saving them as a new scenario in a project |

About generate_test_cases

• Takes a testType, a requirement (the main input — the more specific, the better the result), and optionally affectedPage and additionalCriteria.
• Preview by default. Pass saveToProject ({projectId, scenarioName}) to persist the result as a new scenario instead — without it, nothing is saved anywhere.
• Consumes your real AI-generation quota, the same one the app's UI uses: Basic = 5 generations lifetime (never resets), Advanced = 200/month, Enterprise = 500/month. There's no undo on a Basic-plan call.
• Generation can take a little while; if it's still running after ~90 seconds the tool returns a jobId instead of a result — call it again with resumeJobId set to that value to pick up where it left off (results are only kept for 10 minutes).
• The number of test cases generated (~7) isn't configurable — that's fixed by the underlying platform feature.

1. Create a Personal Access Token

In the Evaficy app https://app.evaficy.com login then go to: Settings → API tokens → Create Token.

• Give it a name you'll recognize later (e.g. "Evaficy MCP - laptop").
• The raw token is shown once, right after creation — copy it immediately, it can't be retrieved again.
• The dialog doesn't currently let you set an expiry from the UI; the token won't expire unless you set one via the API directly. Treat it like a password and revoke it from the same Settings page if it's ever exposed.

2. Configure your client

Set the environment variable in .env file:

• EVAFICY_TOKEN — the token from step 1 (required)

Claude Code

claude mcp add evaficy -e EVAFICY_TOKEN=<your-token> -- npx evaficy-mcp

Cline (VS Code)

Add to Cline's MCP settings file:

{
  "mcpServers": {
    "evaficy": {
      "command": "npx",
      "args": ["evaficy-mcp"],
      "env": {
        "EVAFICY_TOKEN": "<your-token>"
      }
    }
  }
}

Example flow

Once configured, you just talk to your AI client in plain language — it picks the right tool and fills in the parameters. A typical session, going from "what do I have" to "generate and save tests for it":

1. List your projects

List my Evaficy projects.

Calls list_projects. Returns every project you own or belong to, each with its id, your role, and counts (scenarios, members) — this is how you find the project id you'll need in step 2.

2. Generate test cases for one of them

Generate functional test cases for the Checkout page in project <projectId from step 1> (or <Project Name>) requirement: "Users must be able to apply a discount code before payment and see the updated total." Save them as a new scenario called "Discount Code at Checkout".

Calls generate_test_cases with saveToProject set. Returns the ~7 AI-generated test cases (title, steps, expected results, test data) together with the id/name of the scenario that was just created in that project — from here it's immediately visible in the Evaficy app too.

Leave out the "save them as..." part and it returns the same generated test cases as a preview only, without creating anything — useful for reviewing before you decide whether to keep them.

A note on project names vs. IDs: every tool's projectId input needs a real project id, not a name — but you don't have to look that up yourself. Referring to a project by name in your prompt (like "the Sample Project" above) works fine: your AI client resolves it to an id itself, by calling list_projects first (as in step 1) and matching the name, then passing that id to the tool. This happens automatically as part of the conversation — it's not something this server does internally.

Notes

• The six read tools never create, modify, or delete anything. generate_test_cases is the one exception — it always calls the platform's AI generator (consuming quota), and will also create a new scenario if you pass saveToProject.
• Your token is read from the environment only; it is never logged or written to disk by this server.
• A token inherits the permissions of the account that created it.

License

MIT