Cork Protocol Model Context Protocol
Typed MCP gateway for Cork Protocol operations, enabling secure and deterministic interaction with operation core and signing gate.
README
Cork Model Context Protocol server
cork-mcp gives agents and applications a typed interface to Cork Protocol market data and transaction construction.
It exposes tools for:
- discovering Cork pools, flows, whitelists, limit-order markets, orders, and fills;
- verifying market and deployment evidence;
- inspecting and preparing token authority changes;
- preparing, finalizing, simulating, and reconciling Cork actions;
- constructing caller-owned Safe transactions without collecting confirmations;
- preparing limit orders and market deployments; and
- reporting which capabilities are currently callable.
Read results preserve the upstream response bytes and label them as untrusted source observations. Write tools construct artifacts only. The server does not hold signing keys, confirm Safe transactions, or broadcast on behalf of the caller.
Requirements
- Node.js 22
- npm 10.9
The packages are currently consumed from this repository rather than a public package release.
The runnable server has two explicit modes: a no-network fixture mode for transaction construction tests and an opt-in live read-only mode for Phoenix market data. Production write activation requires operator-supplied evidence, policy, credentials, and chain infrastructure; it is not enabled by this repository alone.
Install
git clone https://github.com/Cork-Technology/cork-mcp.git
cd cork-mcp
npm ci
npm run build
Confirm the runtime and repository are healthy:
npm run check:runtime
npm run ci
Try it locally
Run a complete local client/server handshake:
npm run mcp:smoke
This starts the server over standard input/output, discovers its tools, and constructs a deterministic non-broadcast Safe transaction for a synthetic market.
List the synthetic markets available to the Safe demo:
npm run mcp:safe-demo -- --list
Construct a Safe transaction:
npm run mcp:safe-demo
Construct one proposal for each of the six supported Cork action profiles:
npm run mcp:safe-coverage
The coverage command produces sequential, unsigned Safe proposals for
safeUnwindMint, safeDeposit, safeMint, safeUnwindSwap,
safeUnwindDeposit, and safeRedeem. It is a deterministic local encoding
test, not a chain simulation.
Add -- --json to mcp:safe-coverage to inspect all six complete artifacts.
Select a market and inspect the complete artifacts:
npm run mcp:safe-demo -- \
--market synthetic-wsteth-usdc-2027 \
--shares 5000000000000 \
--min-collateral 2000000 \
--nonce 8 \
--json
The Safe demo is fully local. Its chain, markets, addresses, balances, evidence, and Permit2 signatures are fixtures and must not be submitted to a real chain.
Read live Cork market data
Run the opt-in read-only integration test:
npm run mcp:live-read
The command calls https://api-phoenix.cork.tech through the Model Context Protocol server and canonical Node adapter. It tests:
- all, current, and expired pools;
- pool flows and whitelist entries;
- limit-order markets;
- the orderbook; and
- fills.
Live-read mode exposes only capability and read tools. It permits HTTP GET requests to the configured Phoenix origin and does not expose preparation, signing, submission, or broadcast tools. Add --json for a machine-readable result.
Connect a Model Context Protocol client
Build once, then configure the client to start the server directly with Node.js 22:
{
"mcpServers": {
"cork": {
"command": "/absolute/path/to/node-22",
"args": [
"/absolute/path/to/cork-mcp/packages/gateway/dist/dev-server.js",
"--quiet"
],
"cwd": "/absolute/path/to/cork-mcp"
}
}
}
The default server is fixture-only and performs no network requests. Add --live-read before --quiet to enable the public read-only Phoenix tools.
Run the server manually when debugging a client connection:
npm run mcp:dev
The process waits for a client on standard input/output. This is expected. Server diagnostics use standard error so they do not corrupt the protocol stream.
Tool behavior
Every tool has a closed input schema: unknown fields are rejected. The gateway also enforces credential scopes, capability maturity, bounded work, cancellation, and deadlines before invoking a handler.
The main tool groups are:
| Group | Examples | Effect |
|---|---|---|
| Capabilities | cork.capabilities.v1 |
Read |
| Phoenix data | cork.phoenix.pools.list.v1, cork.phoenix.flows.list.v1 |
Read |
| Limit orders | market, orderbook, fill, maker, taker, cancellation tools | Read or construct |
| Authority | inspect, onboard preparation, revocation preparation | Read or construct |
| Cork actions | mint, unwind, repurchase, redeem lifecycles | Construct, simulate, reconcile |
| Market deployment | quote, prepare, simulate, reconcile | Construct, simulate, reconcile |
| Local demos | market list, Safe unwind, six-profile Safe coverage | Fixture only |
A tool being discoverable does not mean that a production capability is active. cork.capabilities.v1 is the source for the current callable state.
Safe boundary
The server may construct calldata, Safe transaction fields, and the Safe transaction hash. Safe message signatures and Safe transaction confirmations remain separate. The caller owns confirmation, submission, receipt storage, and retries.
For a real Safe proposal, the caller must supply current verified market evidence, the exact Safe configuration and nonce, caller-owned authorization artifacts, and a simulation against pinned chain state. The local demo does not substitute fixture values for those inputs.
Packages
| Package | Purpose |
|---|---|
@corkprotocol/operations |
Browser-safe canonical encoders, validators, hashes, and operation state machines |
@corkprotocol/operations-node |
Phoenix, Market Registry, provider, and inventory observation adapters |
@corkprotocol/gateway |
Model Context Protocol transport, routing, controls, local server, and demos |
@corkprotocol/conformance |
Cross-package and public-artifact conformance tests |
The Rust signing gate is under crates/signing-gate and remains independent from the TypeScript operation core.
Development
npm run format:check
npm run build
npm run typecheck
npm run test
npm run ci
Useful focused commands:
npm test --workspace @corkprotocol/operations
npm test --workspace @corkprotocol/operations-node
npm test --workspace @corkprotocol/gateway
npm test --workspace @corkprotocol/conformance
See docs/local-development.md for local modes and troubleshooting, docs/integration.md for integration contracts, and docs/security.md for trust and custody boundaries.
License
This repository is currently unlicensed for external redistribution. See package.json for the recorded package license state.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.