@sixseven-ai/cli

One install, two roles:

• You use sixseven from the terminal to log in and bind a repo to a SixSeven project.
• Claude Code (or Cursor / Claude Desktop) uses sixseven mcp behind the scenes to read your source code, author whitebox testcases, run them, and read evidence — all without leaving the repo.

Same binary, same token, same project binding.

Quick start

1. Get a SixSeven CI token

1. Sign in (or create an account) at https://sixseven-space.sprout.co.id/
2. Open or create a project
3. Go to Integrations → CI/CD Tokens (Project)
4. Type a label (e.g. my-laptop) → click Generate token
5. Copy the token (starts with 67ci_). It is shown only once — save it in your password manager.

2. Install the CLI

npm i -g @sixseven-ai/cli

3. Run the wizard from your app's repo

cd /path/to/your/app
sixseven setup

The wizard will ask for:

• SixSeven platform URL — https://sixseven-space.sprout.co.id (default; just hit Enter). This is the SixSeven backend itself, not the URL of the app you want to test.
• CI token — the 67ci_… you just generated.

It then binds the repo to a project and (optionally) registers the MCP server with Claude Code.

Two URLs, one wizard: the URL of the app you actually want to test (e.g. your staging or production site) is configured later — per testcase or per exploration — from inside SixSeven, not here.

sixseven setup is a 3-step wizard that:

1. Logs you in — prompts for SixSeven base URL + CI token, verifies it, saves to ~/.sixseven/credentials (mode 0600). Skips this step if a valid token is already saved.
2. Binds the repo — lists your projects, pick one, writes .sixseven/config.json. Skips if a binding already exists (asks before re-binding).
3. Registers the MCP server — if the claude CLI is on your PATH, runs claude mcp add sixseven -- sixseven mcp for you. Otherwise prints copy-paste JSON for Cursor (.cursor/mcp.json) or Claude Desktop.

That's it. Open Claude Code and ask "list sixseven features" to confirm.

Workspace dev? If you haven't installed globally, the wizard auto-detects and uses the absolute path (node /abs/path/dist/bin.js mcp) so the MCP registration still works.

Commands

| Command | What | |---|---| | sixseven setup | Guided wizard — start here. Login → bind project → register MCP. | | sixseven login | Just the login step (saves token to ~/.sixseven/credentials) | | sixseven init | Just the bind step (writes .sixseven/config.json) | | sixseven status | Show binding + token validity + feature/testcase counts | | sixseven doctor | Diagnose config, token, backend reachability, project access | | sixseven mcp | Start MCP server on stdio (spawned by Claude Code; you don't run this directly) |

Advanced: manual MCP registration

If you skipped the wizard's MCP step:

# Claude Code
claude mcp add sixseven -- sixseven mcp

# Cursor — add to .cursor/mcp.json
{
  "mcpServers": {
    "sixseven": { "command": "sixseven", "args": ["mcp"] }
  }
}

MCP tools exposed

Whitebox (source code)

| Tool | Purpose | |---|---| | sixseven.read_code | Read a text file inside configured codeRoots | | sixseven.list_dir | List a directory inside the repo | | sixseven.search_code | Substring/regex search across codeRoots (≤200 hits) |

Features & testcases

| Tool | Purpose | |---|---| | sixseven.list_features | List features of the bound SixSeven project | | sixseven.create_feature | Create a feature (group of testcases) | | sixseven.list_testcases | List testcases (optionally per platform) | | sixseven.get_testcase | Fetch a testcase by id | | sixseven.create_testcase | Push a new testcase to a feature | | sixseven.update_testcase | Patch an existing testcase |

Runs, execution, and evidence

| Tool | Purpose | |---|---| | sixseven.list_runs | List recent runs for the bound project | | sixseven.create_run | Create a run for one or more testcases | | sixseven.execute_run | Trigger execution (fire-and-forget, 202) | | sixseven.get_run | Get current run state (status, reportUrl, timestamps) | | sixseven.cancel_run | Cancel an in-flight run | | sixseven.wait_for_run | Poll until terminal status (or timeout). Returns summary + final run | | sixseven.list_run_testcases | List testcases attached to a specific run | | sixseven.get_evidence | Aggregate per-case attempts + step events + failed-step screenshots |

Auto-explore (blackbox)

| Tool | Purpose | |---|---| | sixseven.start_exploration | Kick off SixSeven's BFS crawler on seed URLs (server-side Puppeteer) | | sixseven.get_exploration | Read the latest exploration snapshot (status + counts) |

Bug filing

| Tool | Purpose | |---|---| | sixseven.file_bug | Open an issue report against the bound project (after auto-fix loop confirms a real defect) | | sixseven.list_bugs | List issue reports for the project | | sixseven.get_bug | Fetch a single issue report by id |

Context

| Tool | Purpose | |---|---| | sixseven.whoami | Echo current user + project context |

Security & telemetry

• No telemetry. The CLI makes no outbound calls except to the SixSeven backend you configured in .sixseven/config.json. There is no analytics, no crash reporter, no usage ping.
• Source code stays local. The CLI never auto-uploads files. Source content only leaves the machine if the developer (or Claude Code, on the developer's instruction) explicitly bakes it into a testcase payload.
• File reads are sandboxed. Reads are constrained to configured codeRoots, paths in exclude are refused, and path-traversal attempts (../, absolute paths) are rejected.
• Stdout in sixseven mcp is strict JSON-RPC. Human-readable logs go to stderr. Set SIXSEVEN_DEBUG=1 to include stack traces on errors.
• Credentials. Tokens are saved to ~/.sixseven/credentials with mode 0600. The directory is created with mode 0700. Tokens are never logged.

Walk-through

See USAGE.md for end-to-end prompt examples that drive Claude Code through generate → run → evidence → auto-fix → file-bug loops.

Config shape

.sixseven/config.json:

{
  "baseUrl": "https://sixseven.sprout.co.id",
  "projectId": "uuid-of-project",
  "defaultFeatureId": null,
  "codeRoots": ["src", "apps"],
  "exclude": ["node_modules", "dist", ".git", ".next"],
  "platform": "web"
}