Platform

What sidclawhq/platform does

The sidclawhq/platform MCP server acts as a governance layer in front of another MCP server. Instead of allowing an agent to call upstream tools directly, the proxy intercepts each call and sends it through policy evaluation. The result determines whether the action proceeds, waits for review, or is rejected before the upstream server receives it.

The project describes three possible policy outcomes:

• allow: execute the action immediately and record the decision.
• approval_required: present the action context to a human reviewer, then execute or reject it based on the review.
• deny: stop the action before execution and record the decision.

This makes the proxy suitable for MCP deployments where tool use needs explicit controls, ownership, or an audit record. The repository also provides SDK and integration paths for other agent frameworks, but the MCP proxy is the relevant component when the client communicates through MCP.

How it works

A request follows an identity, policy, approval, and trace sequence. The agent is associated with an owner and scoped permissions, the requested action is evaluated against explicit rules, and higher-risk actions can be held for human review with contextual details. Once the decision is complete, the action either reaches the upstream MCP server or is blocked. The resulting decision is written to a hash-chained audit trail.

The sidclawhq/platform MCP server is intended to sit between an MCP client and an existing MCP server. It does not replace the upstream server's tools; it governs access to them. The README shows examples involving database access, email, production scaling, and clinical workflows, but the proxy is described as wrapping any MCP server rather than being limited to those examples.

Setup and configuration

The documented stdio configuration runs the proxy through npx:

{
  "command": "npx",
  "args": ["-y", "@sidclaw/sdk", "sidclaw-mcp-proxy", "--transport", "stdio"]
}

The configuration also supplies a SidClaw API key and an agent identifier through environment variables. The API key authenticates requests to the governance service, while the agent identifier associates calls with the governed agent. The README's other examples use SIDCLAW_BASE_URL when pointing integrations at a SidClaw API; the proxy configuration excerpt specifically shows SIDCLAW_API_KEY and SIDCLAW_AGENT_ID.

The project can also be deployed as a self-hosted instance. Railway is documented as a deployment option that provisions the API, dashboard, and Postgres together. Vercel is described as a dashboard-only option that must point to an existing SidClaw API.

Tools and capabilities

The proxy provides governance capabilities rather than a separate catalog of business tools. Its documented functions include:

• Intercepting calls made to an upstream MCP server.
• Evaluating calls against policies.
• Allowing, denying, or pausing actions for approval.
• Showing reviewers context for actions awaiting approval.
• Recording decisions in hash-chained audit trails.
• Running through the stdio transport for MCP clients.

The sidclawhq/platform MCP server is documented as compatible with Claude Desktop, Cursor, VS Code, and GitHub Copilot. The broader project also lists integrations for LangChain, OpenAI Agents, Claude Agent SDK, Google ADK, and other frameworks.

Limitations and notes

The provided material does not define the policy language, approval interface configuration, API pricing, or the complete environment-variable list for the proxy. A SidClaw API key and agent identifier are shown in the MCP setup, so this is not a credential-free proxy configuration. The local npx sidclaw-demo command launches preloaded dashboard scenarios for demonstration; it is not the MCP governance proxy itself.

License information is split in the repository: the SDK is marked Apache-2.0, while the platform is marked FSL 1.1. Check the relevant component's license before redistributing or deploying it.

Getting started with this sidclawhq/platform MCP server

Always refer to the official documentation for the most accurate and up-to-date information.