Trusty Squire

What Trusty Squire MCP server does

Trusty Squire MCP server gives coding agents a controlled way to handle service authentication, API credentials, and payments. An agent can complete a SaaS signup, verify an email, create an API key, and place that key in a vault rather than returning it in the conversation. The README uses Pinecone as an example of this workflow.

It also supports sign-in through Google or GitHub on sites that offer those options. For purchases, the agent can request an order while the user approves the transaction; the card number is not provided to the agent. Saved credentials can be used for API calls without exposing the key itself.

How it works

The package installs a local connection layer that links an MCP client with Trusty Squire. After installation, the user signs in with Google or GitHub, restarts the coding agent, and can request operations in natural language. Credential and payment actions are represented as MCP tools, including approval, access management, observation, and completion steps.

The Trusty Squire MCP server keeps API keys in its vault instead of placing them in an .env file or the agent's context. Its credential tools include listing stored credentials, using or fetching a credential, editing credentials, and granting or revoking application access. Payment-related tools include listing payment cards and injecting card details into an approved operation without exposing the card number to the agent.

Setup and configuration

Install and connect it with:

npx -y @trusty-squire/mcp@latest connect

The command lets you choose a coding agent during setup. To select one directly, use a target such as:

npx -y @trusty-squire/mcp@latest connect --target=codex

On Linux, the connector registers and starts a systemd user service. On macOS, it uses a launchd user agent. A working user service manager is required. MCP clients connect to this broker; if it is stopped, clients report that the broker is not running after bounded retries. Installer-created client entries use @trusty-squire/mcp@latest.

Tools and capabilities

The exposed MCP tools cover these areas:

• Operation lifecycle: operate_start, operate_observe, operate_finish, outcome, reason, and error
• Credential access: list_credentials, use_credential, fetch_credential, and edit_credential
• Payment handling: list_payment_cards and inject_card
• Application permissions: grant_app_access and revoke_app_access
• Approval and workflow records: approval_id, audit_log, and operate_recipe_save
• Code-related operations: code

These tools support account setup, OAuth and dashboard workflows, verification steps, API calls using stored keys, and approved purchases.

Limitations and notes

The material describes support for Claude Code, Codex, Cursor, OpenCode, Goose, Cline, Continue, and Hermes. Of those named clients, this listing maps to Claude Code, Codex, Cursor, Cline, and Continue. OpenCode, Goose, and Hermes are not included as compatible-client slugs here.

A user service manager must be available on the host. The README does not specify environment variables, a license, or detailed vault-retention behavior. It also does not identify a fixed catalog of SaaS integrations beyond the Pinecone signup example and the whitejade.xyz purchase example.