IDE Agent (Cursor / VS Code)

The @testflowkit/mcp server connects your IDE to the step catalog and project config, so AI-generated tests use only registered sentences.

Setup

1. Initialize the project:

tkit init

2. Add MCP config:

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "testflowkit": {
      "command": "npx",
      "args": ["-y", "@testflowkit/mcp"]
    }
  }
}

VS Code (.vscode/mcp.json):

{
  "servers": {
    "testflowkit": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@testflowkit/mcp"]
    }
  }
}

3. Prompt the agent:

"Write a test for user registration using the available API operations."

The agent reads the step catalog and testflowkit.yml, then drafts scenarios with valid steps only.

Requirements

tkit CLI installed (for catalog version detection)
Node.js >= 22
testflowkit.yml in the project root

MCP tools

Tool	Purpose
`get_step_catalog`	Full sentence list
`search_sentences`	Filter by keyword/category
`read_test_config`	APIs, pages, elements (secrets redacted)
`list_features` / `read_feature`	Browse feature files
`write_feature` / `write_macro`	Create or update `.feature` files
`get_guidelines`	Macro and agent authoring docs

Agent config in testflowkit.yml

agent:
  default_tags_for_draft: "@wip @ai-generated"
  run_command: "tkit run --tags @wip"

The agent: block is ignored by tkit run — only the MCP server reads it.

Missing sentences

If no registered step matches your intent, the agent reports a missing_sentence block instead of inventing one. Use these to request new steps via GitHub issues.

Troubleshooting

Problem	Fix
MCP won't start	Run `npx @testflowkit/mcp` manually; check Node >= 22
Catalog empty	Verify `tkit version` works
Wrong steps used	Ask the agent to call `get_step_catalog` first

Next Steps

Step Catalog — Browse all available sentences
testflowkit.yml — Project configuration