About kit
Toreva kit provides MCP, SDK, and CLI connectors for developers using their own compute with Toreva governed objects. Its MCP tools support policy-bounded wallet actions, unsigned transaction preparation, and receipts for outcomes; no money moves through the connector.
Key features
- 01Read or scan main and agent wallets
- 02Set hard limits on agent wallets
- 03Simulate actions before preparation
- 04Prepare unsigned transactions
- 05Explain actions and retrieve receipts
- 06Return refusal receipts for unsafe actions
Use cases
- 01Inspect a main or agent wallet
- 02Test a proposed action before preparing it
- 03Prepare an unsigned transaction within set limits
- 04Review the receipt for an accepted or refused outcome
MCP support
The MCP server exposes tools for wallet reads, action simulation, unsigned transaction preparation, explanations, and receipts. It connects to Toreva through the gateway relay, returning a receipt for accepted, pending, or refused outcomes.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project README
toreva kit
Connector primitives for Toreva governed objects. Bring your own compute, use the gateway relay, and require receipts for every outcome.
Kit provides SDK, CLI, and MCP connectors for developers connecting their own compute to Toreva governed objects. The setup paths below are developer integrations.
The Day-1 Mac helper scope is Pair and Unpair only, for every user. It must request no Screen Recording or Accessibility permissions and take no jobs. Developer connector setup is separate from helper pairing.
The front door is the object store. MCP clients, SDK callers, CLI users, and customer-owned agent fleets all read and write the same Toreva objects through the gateway relay, and every accepted, pending, or refused outcome returns a receipt.
GitHub can prove which organization and admin are acting. It is not the transport for Toreva work messages. Compute is bound per scope, and the live paths put model execution on the developer's own account.
No money moves through this connector.
The Three Doors
Door 1: MCP
For developer integrations, start here. Your AI client connects the Toreva MCP server, your own client subscription does the work, and Toreva receives only the object reads and writes you authorize.
npx toreva init --client=claude-desktop
npx toreva login
npx toreva doctor
Restart the client, then ask it about your Toreva objects or receipts. Supported
client config targets are claude-desktop, cursor, and openclaw.
Door 2: Customer-Owned Fleet
Use this when you need unattended or scheduled agents. A GitHub organization
admin runs agents in their own org, on their own runners, with their own model
credentials. The agents call Toreva through MCP, SDK, or CLI and land answers
back on Toreva objects with receipt_id proof.
Start here: GitHub Org Agent Fleet.
Door 3: Hosted Compute
Hosted compute means Toreva runs the agents and carries the marginal model cost. This door is deferred and is not documented as available. Use MCP or a customer-owned fleet today.
How Answers Land
The object store is the receiving system. A work surface prepares a relay
request, sends it to https://gateway.toreva.com/relay, and treats the result
as accepted only when the gateway returns a receipt_id.
For the shared install, invoke, and receipt shapes, see Work-Surface Connector Primitives.
MCP Tools
| Tool | What it does |
|---|---|
toreva_establish_agent_wallet | Create the agent wallet with hard limits |
toreva_read_or_scan | Read or scan a main wallet or agent wallet |
toreva_simulate_action | Test an action before anything is prepared |
toreva_prepare_unsigned_transaction | Prepare an unsigned transaction after simulation |
toreva_explain_action | Explain an action or receipt |
toreva_get_receipt | Fetch one receipt |
toreva_refuse_action | Refuse an unsafe action and return a refusal receipt |
Install
The local package runs over stdio. Use either the token written by
toreva login or an explicit RELAY_AUTH_TOKEN from your own secret store.
npm install -g @toreva/mcp
toreva login
toreva-mcp
For runner-owned environments:
RELAY_AUTH_TOKEN=your_token toreva-mcp
For local development:
pnpm install
pnpm --filter @toreva/mcp build
RELAY_AUTH_TOKEN=your_token node packages/mcp/dist/index.js
Package
@toreva/mcp is the Rung-1 MCP package. Version 0.2.0 is stdio-local only.
The public metadata is server.json and packages/mcp/package.json.
Packages
| Package | What |
|---|---|
@toreva/mcp | MCP server for agent integration |
@toreva/types | Shared schemas and types |
@toreva/sdk | TypeScript client library |
@toreva/cli | Command-line interface |
Verify
pnpm test
pnpm --filter @toreva/mcp build
Regulatory notice
This software provides tooling for interacting with the Toreva policy service. It does not provide financial advice, investment advice, trading advice, or any other form of advice. Use of this software does not create a fiduciary relationship, advisory relationship, or any other professional relationship between you and Toreva Pty Ltd.
Toreva Pty Ltd is not responsible for modifications made to this software by third parties, including modifications that alter or remove compliance language, disclaimers, or risk warnings. If you use a modified version of this software, you do so at your own risk and are responsible for ensuring your use complies with applicable law.
License
MIT — see LICENSE
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.