MCP Tool Reference

Conduit exposes 54 tools through the MCP server, organized into eight categories. Each tool is callable by any connected AI agent.

Terminal Tools

• TerminalExecute — Execute a command in an active SSH session. Returns the command output.
• TerminalReadBuffer — Read the terminal buffer with a configurable number of history lines.
• TerminalSendKeys — Send keystrokes to the terminal, useful for interactive prompts and control sequences.
• TerminalCreateLocal — Create a local shell session (bash/zsh) without an SSH connection.

RDP Tools

• RdpScreenshot — Capture a screenshot in PNG or JPEG format with quality control and optional regional capture.
• RdpClick — Click at coordinates with support for left, right, and middle buttons, as well as single and double-click.
• RdpType — Type text into the active RDP session.
• RdpSendKey — Send key press, key down, or key up events.
• RdpMouseMove — Move the mouse cursor to specific coordinates.
• RdpDrag — Drag from one point to another.
• RdpScroll — Scroll up or down at the specified coordinates.
• RdpResize — Resize the RDP display dynamically via the RDPEDISP channel.
• RdpGetDimensions — Query the current session dimensions.

VNC Tools

• VncScreenshot — Capture a screenshot of the remote VNC session.
• VncClick — Click at coordinates.
• VncType — Type text into the VNC session.
• VncSendKey — Send key events.
• VncMouseMove — Move the mouse cursor.
• VncDrag — Drag between two points.
• VncScroll — Scroll at the specified coordinates.
• VncGetDimensions — Query the current session dimensions.

Web Tools

Web sessions support 21 tools across three interaction models: coordinate-based (matching RDP/VNC), DOM-aware (web-exclusive), and tab management. Coordinates are specified in screenshot image space and automatically scaled to viewport CSS pixels.

Content & Navigation

• WebScreenshot — Capture a screenshot of the web session. Returns a base64-encoded image.
• WebReadContent — Read page content by CSS selector. Supports text, HTML, or markdown output.
• WebNavigate — Navigate the web session to a URL.
• WebGetDimensions — Query the current viewport dimensions of the web session.

Coordinate-Based Interaction

• WebClick — Click at coordinates with support for left, right, and middle buttons, as well as single and double-click.
• WebType — Type text at the currently focused element in the web session.
• WebSendKey — Send key press, key down, or key up events with optional modifier keys.
• WebMouseMove — Move the mouse cursor to specific coordinates.
• WebDrag — Drag from one point to another.
• WebScroll — Scroll up or down at the specified coordinates.

DOM-Aware Interaction

• WebClickElement — Click an element by CSS selector. Uses DOM click() so it works even if the element is off-screen.
• WebFillInput — Fill an input field by CSS selector. Sets the value with native setter and dispatches input/change events for React, Vue, and Angular compatibility.
• WebGetElements — Discover interactive elements on the page (buttons, links, inputs, selects) with their text, CSS selector, and bounding box.
• WebExecuteJs — Execute JavaScript code in the page context and return the result.

Tab Management

• WebListTabs — List all tabs in the session with their URLs, titles, and active status.
• WebCreateTab — Open a new tab, optionally navigating to a URL. Maximum 12 tabs per session.
• WebCloseTab — Close a specific tab by its ID.
• WebSwitchTab — Switch the active tab in the session.
• WebGoBack — Navigate the active tab backward in its browsing history.
• WebGoForward — Navigate the active tab forward in its browsing history.
• WebReload — Reload the active tab.

Entry Tools

Entry tools provide read access to vault entry metadata. Read operations automatically redact fields marked as !!secret!! in the vault, replacing them with [REDACTED] to prevent AI agents from leaking visually-masked secrets.

• EntryInfo — Get metadata for any vault entry (connection, document, or command). Optionally include the entry's markdown notes.
• EntryUpdateNotes — Update the markdown notes on any vault entry. Requires the agent to show proposed content and receive user approval before calling.

Document Tools

Document tools let AI agents read and write markdown documents stored in the vault. Like entry tools, read operations redact !!secret!! values as [REDACTED].

• DocumentRead — Read the markdown content of a document entry.
• DocumentCreate — Create a new markdown document entry in the vault. Accepts a name, content, and optional folder and tags.
• DocumentUpdate — Update the content (and optionally the name) of an existing markdown document entry.

Credential Tools

• CredentialList — List all credentials stored in the vault. Includes a has_totp flag indicating whether TOTP is configured on each credential.
• CredentialCreate — Create a new credential entry in the vault.
• CredentialRead — Read credential details. Requires explicit user approval.
• CredentialDelete — Delete a credential entry from the vault.

Connection Tools

• ConnectionList — List all connection entries configured in Conduit.
• ConnectionOpen — Open a connection (SSH, RDP, or VNC) by its entry ID. For SSH connections, an optional ssh_auth_method parameter ("key" or "password") overrides the configured auth method preference for that session.
• ConnectionClose — Close an active connection session.