Hycli CLI
Hycli turns a website URL and a goal into a typed CLI and MCP tools that run in your own browser session, with no cookie leaks.
- 451 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Uses your browser sessions and, once enabled, calls write actions in real accounts.
- When a site blocks requests it switches to in-page requests or a local Chromium, which may breach service terms.
Install
- Works in
Manual install
{"mcpServers": {"hycli": {"command": "hycli", "args": ["mcp"]}}}The minimal README example. Better to take the generated settings from the app: the Coding agents section preserves the path and `HYCLI_DATA_DIR`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Uses your browser sessions and, once enabled, calls write actions in real accounts.
- When a site blocks requests it switches to in-page requests or a local Chromium, which may breach service terms.
This is third-party code. Review the repository files before installing.
What it does
Hycli takes a website address and a goal, inspects documentation, frontend bundles and live browser network traffic, then builds a typed CLI and a set of MCP tools for that site: reading, searching, creating and editing records. Action definitions live in a local spec, and a dashboard runs at `http://127.0.0.1:4318`. Cookies, tokens and keys stay in a local vault and never enter model prompts: the agent sees only account labels and redacted response schemas. Write actions are off by default, and the agent asks the user once per CLI whether to enable them. The repository ships a `skills/hycli/SKILL.md` skill with agent instructions.
Who it is for. Developers and analysts automating web services that lack a public API who want to hand those actions to an agent under control.
Good fit when
- A service has no API and routine actions in it should go to an agent.
- An agent in Cursor or Codex must search and draft records in your working account.
- A site spec should be checked without a model through `hycli check`.
- One set of sites and accounts is needed in both the terminal and the dashboard.
Not a fit when
- The service already has an official API or MCP server.
- Site terms forbid automating an account.
- Importing a browser session into a local vault is not acceptable.
Example request
Prepare a CLI for the notes site with the goal of searching and saving drafts, list the available actions and find notes matching release notes.Limitations
Preparing a site needs a key from one of the AI providers (OpenAI, Anthropic, xAI or Z.ai) or a local Codex login. An agent app subscription to Claude or ChatGPT does not replace it. The README lists release 0.1.1, so the project is young. When a site blocks requests, Hycli switches to in-page requests or a local Chromium, and that workaround may breach the service terms. Enabled write actions run immediately, so verify results on the site.
How to disable. Remove the `hycli` entry from the client MCP configuration and revoke writes with `hycli writes SITE`. Sessions and logs sit in the Hycli data directory, which is deleted separately.
FAQ
Which tools does the MCP server expose?
`hycli_prepare`, `hycli_writes`, `hycli_run`, `hycli_check`, `hycli_result` and `hycli_activity`. There is no tool for reading credentials.
Can the agent change account data on its own?
Only after your consent. A new CLI starts with `writes: false`, and during preparation Hycli never executes write requests.
Where do cookies and tokens go?
They stay in the local vault, either the OS keyring or a file with `0600` permissions. Only labels and redacted schemas reach the model.
How do I check that a login is still valid?
Run `hycli check SITE`. It replays verified read chains without a model and separates a signed-out account from a changed site.
Which AI providers does site preparation need?
A key from OpenAI, Anthropic, xAI or Z.ai, or a local Codex login. A subscription from an agent app does not work.
Discussion
No questions or reviews yet.
Related
CLI toolsA plugin and CLI catalog for agents: generates command-line interfaces for GUI apps like GIMP and Blender and installs ready ones via CLI-Hub
Bitrix24 as mission control for AI dev agents: a kanban task conveyor, shared memory and one-button phone acceptance
One advai command for local skills, Chrome extension browser control, external CLIs and a terminal chat with a model
An MCP server and CLI for the BOSS job platform: search vacancies, filter by conditions and shortlist with JSON output
SwitchBot smart-home CLI for devices and scenes, with a built-in MCP server for Claude Code, Codex and Gemini CLI.
MCP server where an agent asks a human questions and waits for answers, which arrive through MCP tools or an HTTP API.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Hycli (this card) | no | 451 | |
| OpenCLI | no | 30,936 | turns a site into a command through ready adapters; Hycli studies the chosen site itself and builds a CLI with checked reads and manual write enabling |
| Unbrowse | yes | 903 | works with sites by capturing and replaying requests; Hycli keeps a spec, separates secrets from the model and asks about writing once per CLI |
README in short
All site actions live in a local spec. Reads run at once, while create, edit and delete are enabled once per CLI.