TOOLBRAID

Browser tools

Connecting

BUILD · CHECK · REPEAT

Developer workspace

Local

Use your existing AI session. Keep page access scoped, capture a reproducible bug, and check site-provided WebMCP tools.

Connection checklist

Read-only checks. No model request or account sign-in.

    Project access Off

    An additional restriction on connected browser pages. Unlisted origins are blocked while a profile is active. This does not sandbox companion file or desktop tools.

    Exact origins, one per line, including the port if used. No paths or wildcards. Native site-tool execution requires read and write access because its effects cannot be verified in advance.

    Loading project settings.

    Capture a bug report

    Start on the connected top-level page, reproduce the issue without navigating, then stop. Captures console errors and failed request metadata for up to 5 minutes. It does not collect headers, cookies or request/response bodies. Close DevTools first.

    Adding a screenshot captures unredacted visible pixels. Text filtering is best-effort, not a privacy guarantee. Inspect the entire preview before sharing.

    Capture is off. Nothing is sent to an AI.

    Attach the JSON file to your existing AI conversation. The draft stays in memory until discarded or the extension worker stops; downloaded files remain until you delete them.

    WebMCP regression checks

    Save exact inputs and assertions, then compare runs on the same site. Discovery makes no tool calls. A run needs a separate review; it is never retried automatically.

    Assertions use JSON Pointer paths, for example /count. Operations: equals, exists, type, error. Inputs and expected values are stored locally; do not save secrets. Review site tools yourself before marking a step readOnly.

    No scenario selected.

    Read only what changed

    Your MCP client can call toolbraid.context.read_changes and return its cursor on the next read. The first response is a bounded snapshot; later responses contain changes or an unchanged result. Navigation, expiry or an invalid cursor requires a fresh snapshot. Savings depend on the page and tokenizer; no fixed reduction is promised.

    OPTIONAL CHAT

    Your browser agent

    ChatGPT account

    You can use ToolBraid from your existing MCP client instead. ChatGPT sign-in is not required to connect a page or check the companion connection.

    Chat through Codex using your ChatGPT account and its Codex limits. No API key or separate API billing. Requires the Windows companion and Codex signed in on this PC.

    Connection & model
    Codex executable

    Not signed in? Run codex login and choose your ChatGPT account, then connect here. ToolBraid never asks for your password or copies your login tokens.

    Connect to start a conversation.

    Ask about the page, inspect native WebMCP tools, or catch up with your X community.

    A run stays on its selected tab even when you switch tabs. Chat asks before changes. Up to 40 recent messages are kept locally and reused as conversation context; clearing removes this local copy. Shared content is sent to OpenAI, subject to your account settings.

    Agent activity

      STAY CONNECTED. KEEP BUILDING.

      Your X community

      Review visible replies and mentions with their original links. Ask the agent for a summary or a draft, then decide what to send. No automatic replies or DMs.

      Optional quiet monitoring

      Open X Notifications, Mentions or a post conversation in a dedicated tab. Keep Edge and that tab open. ToolBraid can refresh it in the background and notify you only about newly observed posts. It does not call a model or consume Codex usage while monitoring.

      Monitoring is off. Only currently rendered posts can be inspected; this is not a complete account archive.

      SITE-PROVIDED TOOLS

      WebMCP enabled

      Discover and call tools registered by compatible sites through the browser's native WebMCP API. Availability depends on the browser and site; ToolBraid's other browser tools remain available independently.

      Connect a site to check native WebMCP availability.

        Current workflow

        NOW Connecting to the page
        NEXT Discover live tools
        REQUIRES YOU Nothing yet

        LOCAL CONTROL PLANE

        AI control

        Active

        Direct execution is enabled under the local owner policy.

        Control planeLocal
        Agent modelLocal / cloud via MCP

        The integrated chat uses your ChatGPT account through Codex. External MCP clients choose their own models. CAPTCHA, MFA, and passkey steps remain explicit handoffs.

        ACTIVE CONTEXT

        Current page

        Waiting

        No active HTTP(S) page

        Activate ToolBraid on a tab to inspect its live context.

        Tab
        —
        Snapshot fingerprint
        —

        MULTI-PAGE CONTINUITY

        Missions 0

        This bounded brief travels with the mission so every page keeps the same intent.

        No active mission. Start one from the current page.

        Each mission keeps exact tab, frame, session, origin, and page-fingerprint ownership.

        INTERVENTION HANDOFF

        CAPTCHA / MFA steps 0

        No CAPTCHA, MFA, or passkey step is waiting.

        Credentials stay inside the approved site. ToolBraid stores no password, one-time code, or raw login URL.

        DISCOVERY

        Available tools 0

        No quarantined tools
        Selected packs 0
        Active packs 0
        Tool budget 0 / 0

        No tools discovered yet.

        MULTIMODAL CONTEXT

        Evidence 0

        0 analyzed · 0 degraded · 0 blocked

        No visual, audio, or video evidence on this page.

        Visible-tab capture is ephemeral and never grants action authority.

        OPTIONAL ANALYSIS

        Multimodal provider

        Metadata only

        Disabled by default. Enabling requests access only to the exact endpoint origin; the API key stays in extension session storage.

        CONTROL SURFACE

        Page actions 0

        No page actions discovered yet.

        FALLBACK & HISTORY

        Approval records 0

        Current context

        0 actionable

        No approval is actionable on this page.

        History & other contexts

        0 records

        No approval history.

        Owner Autonomy dispatches policy-covered effects directly. Items appear here only when manual fallback is required or for history.

        EXECUTION PROOF

        Receipts 0

        Checking chain

        No action has been dispatched.

        Audit trail
        Audit head —

        Audit entries appear after activation.