Faithful control of your real, logged-in Chrome profiles: MCP server + MV3 extension + authenticated localhost bridge. No copied profiles, no debug profile, no remote-debugging port, no Edge.
Chrome Faithful defines 9 tools with explicit schemas and descriptions. Naming is action-oriented (chrome_profiles, chrome_tabs, chrome_cdp) and mostly clear. Descriptions are present and detailed (avg ~180 chars), explaining WHAT each tool does and WHEN to use it. However, parameter descriptions are sparse or missing entirely, most params lack inline documentation of constraints, formats, or dependencies. Schemas are well-formed JSON Schema with types and enums, but lack descriptions for individual parameters. Error handling guidance is minimal; tools do not explain recovery paths or categorize failures. Output schemas are not documented. The server demonstrates solid foundational quality but falls short of production-grade polish in parameter clarity and error guidance.
Use raw CDP or bounded Network projections in one exact profile. action=send is an unrestricted, fully trusted CDP capability and may return authenticated page content, cookies, storage, tokens, URLs, and headers. readEvents always redacts Network headers, query strings, and post data. Response/request projection actions keep raw bodies inside the plugin and reject sensitive selected fields.
Read or write the Chrome profile clipboard through the authorized extension offscreen document.
Search Chrome history inside one exact profile.
Use the Codex-compatible Playwright locator surface in an exact Chrome profile. For editable form controls, prefer fill/type/press here: these actions scroll first, perform real tab-scoped CDP mouse and keyboard input, and preserve framework input state. Do not replace textContent with Runtime.evaluate for form entry.
List existing local Google Chrome profiles without launching Chrome. Returns only directory and display names.
Parameter descriptions missing or minimal. Most params (profileName, tabId, action, etc.) lack inline documentation of expected format, constraints, or dependencies. LLMs cannot infer whether tabId accepts string or number, or what values 'action' enum accepts without reading the schema.
Output schemas not documented. Tool descriptions do not specify what fields are returned, their types, or structure. Agents cannot plan downstream calls or extract required IDs (e.g., does chrome_tabs return tab_id or tabId?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Securely bootstrap and start one exact existing local Chrome profile with ordinary Chrome, then wait for that profile's installed extension ID and profileName to register. Never returns the one-use token or secret and never uses remote debugging or copied user data.
List exact connected Chrome profiles. Always select by metadata.profileName.
Run tabs and Runtime.evaluate checks inside one exact Chrome profile. If tabId is omitted or names an inaccessible browser-internal page, the plugin creates an about:blank task tab and always closes it before returning.
List, get, create, activate, navigate, reload, go back/forward, or close tabs in one exact profile.
Error handling lacks recovery guidance. Tool descriptions do not explain what errors can occur, how to categorize them (retryable vs. fatal), or what the LLM should do next. E.g., chrome_profile_start timeout, should the agent retry, increase timeoutMs, or try a different profile?
Parameter dependencies undocumented. chrome_cdp 'action' enum determines which params apply (send uses 'method'/'params'; readEvents uses 'options'). chrome_tabs 'action' enum determines whether 'tabId' or 'url' is required. These dependencies are not stated in parameter descriptions.
Destructive operations lack confirmation pattern. chrome_tabs 'close' action and chrome_clipboard 'write' action modify state but do not offer dry-run or confirmation steps. Agents may accidentally close tabs or overwrite clipboard without explicit user approval.