Tether WDK CLI Adds MCP Wallet Access for Claude, OpenClaw and Local Agents
Tether's Wallet Development Kit now has a beta command-line wallet and bundled Model Context Protocol server that can give compatible AI clients structured access to a locally managed self-custodial wallet. The current @tetherto/wdk-cli release documented by Tether is 1.0.0-beta.3, requires Node.js 22.18.0 or later, and installs wdk, wdk-daemon and wdk-mcp binaries.
The bundled MCP server exposes 11 tools for network and token discovery, addresses, balances, transaction history, chain-specific methods, token transfers and MoonPay buy/sell flows. Tether provides automated MCP setup for Claude Desktop, Claude Code and OpenClaw, while other MCP-compatible clients can connect over stdio.
The security model matters because an unlocked WDK CLI wallet behaves like a local hot wallet for the duration of its unlock session. On Unix-like systems, another process running as the same OS user and able to reach the owner-only daemon socket can request signing or sending without another passphrase prompt. Tether's documentation recommends dedicated development wallets, limited funds and finite unlock time-to-live values for agent integrations.
What ships in WDK CLI beta.3
WDK CLI combines a human-facing command line with a local daemon and an MCP interface over the same wallet store.
| Component | Role |
|---|---|
wdk |
Wallet, read, send, configuration, network, token, fiat-ramp and MCP setup commands |
wdk-daemon |
Holds unlocked wallet instances and handles local wallet requests |
wdk-mcp |
Exposes selected WDK operations as MCP tools |
The CLI supports wallet creation/import/export, address derivation, balances and history, native and registered-token transfers, custom networks and tokens, chain-specific module methods, and MoonPay on/off-ramp URL creation. Tether documents 37 leaf CLI commands in beta.3.
Installation is currently:
npm install -g @tetherto/[email protected]
Tether specifically identifies the scoped @tetherto/wdk-cli package as the official CLI; the similarly named unscoped npm package is a different package.
The 11 MCP tools
The bundled server deliberately exposes a smaller surface than the full CLI.
| MCP tool | Wallet required | Purpose |
|---|---|---|
get_networks |
No | List supported networks |
list_tokens |
No | List registered tokens |
get_token |
No | Read a token registry entry |
get_address |
Yes | Derive wallet addresses |
get_balance |
Yes | Read balances, including aggregate native balances |
get_history |
Yes | Query indexer-backed transfer history |
list_methods |
No | Discover chain-specific module methods |
call_method |
Yes | Invoke a declared module method |
send_token |
Yes | Preview or broadcast a transfer |
buy_crypto |
Yes | Create a signed MoonPay buy URL |
sell_crypto |
Yes | Create a signed MoonPay sell URL |
Wallet administration and persistent configuration remain outside this bundled MCP surface. An MCP client cannot use these 11 tools to create, import, export, unlock, lock, delete or rename a wallet, or change persistent network, token and module configuration.
That boundary applies to the MCP server itself. A local AI application that separately has shell access may still be able to invoke CLI commands available to its operating-system user.
Claude Desktop, Claude Code and OpenClaw setup
Tether provides setup commands for three clients:
wdk mcp setup --ai-tool claude-desktop
wdk mcp setup --ai-tool claude-code
wdk mcp setup --ai-tool openclaw
Claude Desktop receives a wdk-wallet MCP server entry in its platform-specific configuration. Claude Code setup uses its MCP command-line registration. OpenClaw receives an MCP server configuration pointing to the installed WDK MCP script.
The wallet must be unlocked before wallet-dependent MCP calls can execute. Tether's example uses a short five-minute session:
wdk wallet unlock --name agent-dev --ttl 5
The TTL is absolute and MCP activity does not extend it. Setting --ttl 0 disables automatic expiry, which materially increases the period during which local processes can use an unlocked wallet.
Security boundary: unlock once, then agents can transact
The CLI stores encrypted wallet seed material under its local configuration directory. On Unix-like systems, the daemon uses an owner-only Unix socket; Windows uses a named pipe. The MCP client sends operation parameters to wdk-mcp, while seed phrases and wallet passphrases are not MCP tool inputs.
The consequential boundary appears after a wallet is unlocked. Tether states that read, signing and send operations during that session do not ask for the passphrase again. Owner-only socket permissions isolate different OS users, while processes running as the same wallet owner can potentially reach the daemon.
For development and evaluation, a practical deployment profile is therefore:
- create a dedicated agent wallet rather than exposing a primary wallet;
- fund it only to the amount needed for the test or workflow;
- use a finite, short unlock TTL;
- constrain the AI client's shell and local-process permissions separately from MCP permissions;
- lock the wallet when the workflow finishes.
These controls follow directly from the documented local authorization model and reduce the value exposed during an agent session.
Bundled MCP server vs MCP Toolkit
Tether also maintains a separate WDK MCP Toolkit, currently documented as v1.0.0-beta.1. It is intended for developers building their own MCP server in application code rather than using the fixed CLI tool surface.
The Toolkit exposes 35 built-in tools, supports 13 built-in chains, and covers additional functions such as swaps, bridges and lending. Its documented write-tool model uses MCP elicitations for explicit user approval before broadcasting transactions.
That creates two distinct deployment choices:
| Option | Best fit | Authorization model |
|---|---|---|
WDK CLI + bundled wdk-mcp |
Local CLI users and agents needing a ready-made wallet interface | Human unlock starts a timed session; subsequent wallet operations can use the unlocked daemon |
| WDK MCP Toolkit | Applications building a customized agent-wallet server | Programmable tool selection; documented write operations use human confirmation |
Developers who need cumulative spending limits, application-specific approval logic or a narrower custom tool set should implement those controls at the application/toolkit layer rather than infer them from the CLI unlock session.
Deployment assessment
WDK CLI beta.3 makes agent-to-wallet integration unusually direct: one local wallet store can serve human CLI commands and structured MCP calls without sending the seed or passphrase through the MCP request path. The trade-off is a clear local trust boundary. Once unlocked, the wallet is intentionally usable by the local daemon for the session, so OS-user isolation and the agent's other capabilities become part of the wallet's security perimeter.
For experimentation, test wallets and bounded automation, the CLI offers the shortest path. Applications handling meaningful balances should evaluate the MCP Toolkit or an application-specific approval layer, isolate the agent runtime, and enforce transaction policy independently of conversational prompts.