Mattheus MCP server
Connect Claude or another MCP client to Mattheus tools for current-launch research, monitoring, and approval-first workflows where enabled.
Short answer
Configure an MCP client with $MCP_SERVER_URL and an mt_ Bearer token. Execution tools are disabled by default with MCP_PUBLIC_EXECUTION_TOOLS_ENABLED=false, and live service tool-count claims require pnpm smoke:mcp:live evidence before publication.
MCP setup
Configure the Streamable HTTP endpoint with an mt_ Bearer token and verify live tool availability before publishing tool-count claims.
Assistant boundaries
Assistant tools can research, draft, and preview. Execution-capable tools are disabled by default and remain approval-first, policy-gated, referenceId-based, launch-mode-aware, and constrained by the authenticated Agent Portal API.
Provenance and freshness
Successful MCP results include redacted structured provenance and freshness metadata. If an upstream response does not report asOf or stale state, those fields remain null and the freshness status is not_reported.
Who This Is For
Developers connecting AI assistants, internal agents, or MCP-compatible clients to Mattheus account-aware tools.
Before You Begin
- An MCP-compatible client
- An mt_ API key
- Passing MCP live smoke evidence before publishing production tool-count claims
- Execution approval and policy gates for any execution-capable tool
Answer engine brief
Mattheus overview
Mattheus is a private AI trading workspace for researching markets, testing reproducible strategies, reviewing portfolio context, and preparing governed actions. Developers can connect through REST APIs, SDKs, and MCP while account authority, policy checks, and approvals remain separate.
Does Mattheus require approval before execution?
Yes. An interactive request requires an exact short-lived one-time approval confirmed with fresh wallet reauthentication. API-key automation requires a separately activated, scoped agent session. Both paths still require wallet or signing readiness and policy checks.
Which markets are documented?
The active launch documentation covers Hyperliquid market data, paper trading, and gated Live workflows. Each page states whether an example is read-only, simulated, paper, or eligible for approval-gated Live execution.
How does policy-gated execution work?
Execution-capable requests are preview-first and must pass wallet readiness, protocol certification, feature gates, idempotency, configured policy controls, and either an exact interactive approval or a bounded API-key agent session before any live action.
Can users revoke permissions?
Users should be able to revoke supported scoped permissions and API keys. Documentation should keep stop, cancel, revoke, and policy rejection paths clear before users approve live execution.
Example
What this example does
Follow this example in order. Read the expected result below before connecting it to a live account or execution-capable workflow.
{
"mcpServers": {
"mattheus": {
"url": "$MCP_SERVER_URL",
"headers": {
"Authorization": "Bearer mt_YOUR_API_KEY"
}
}
}
}Common errors
- 401 Unauthorized: the MCP client did not send Authorization: Bearer mt_...
- Connection failed: confirm the MCP client supports Streamable HTTP and the configured URL is reachable.
- Tool unavailable: production tool availability must be verified with pnpm smoke:mcp:live.
- ACTION_CANARY_ONLY: the action is canary-only and requires beta/canary access.
- POLICY_DENIED: subscription, approval, wallet readiness, agent-session, quota, or configured policy controls blocked execution.
Safety note
MCP can help research, draft, and preview actions where capabilities allow. It does not expose the interactive approval creation or decision routes, and it cannot turn an API key into an approval. Any enabled execution tool still requires an active scoped Agent Portal session and cannot bypass launch mode, feature gates, protocol certification, regulatory gates, signer readiness, or policy controls.