@sidclaw/mcp-tools

@sidclaw/mcp-tools

MCP server that exposes SidClaw governance as callable tools — the complement to the transparent MCP proxy shipped with @sidclaw/sdk.

Where the proxy wraps an upstream MCP server and gates its tools, this server provides governance primitives (evaluate, record, approve, policies) that any MCP-aware agent can call directly. Use it with Claude Code, Claude Desktop, or any MCP host.

MIT licensed.

Install

npx @sidclaw/mcp-tools --version

Configure (Claude Code / Claude Desktop)

Add to your MCP config:

{
  "mcpServers": {
    "sidclaw": {
      "command": "npx",
      "args": ["-y", "@sidclaw/mcp-tools"],
      "env": {
        "SIDCLAW_BASE_URL": "https://api.sidclaw.com",
        "SIDCLAW_API_KEY": "ai_your_key_here",
        "SIDCLAW_AGENT_ID": "claude-code"
      }
    }
  }
}

Restart your MCP host. The agent can now call:

Tool	What it does
sidclaw_evaluate	Pre-action policy check. Returns allow / approval_required / deny.
sidclaw_record	Log an action outcome (success/error + token usage).
sidclaw_approve	Block until a human decides on a flagged action.
sidclaw_policies	List active policies.
sidclaw_session_start	Register a session.
sidclaw_session_end	Close a session.

Plus three resources:

• sidclaw://policies — active rules
• sidclaw://status — instance health
• sidclaw://agent/{agent_id}/history — last 50 traces for an agent

Environment

Variable	Required	Default
SIDCLAW_BASE_URL	yes	—
SIDCLAW_API_KEY	yes	—
SIDCLAW_AGENT_ID	no	claude-code
SIDCLAW_MCP_NAME	no	sidclaw-mcp-tools

When to use this vs the MCP proxy

Scenario	Pick
Wrap an existing MCP server (Postgres, filesystem, GitHub)	@sidclaw/sdk → sidclaw-mcp-proxy
Give the agent explicit governance tools to call	@sidclaw/mcp-tools (this)
You want both — agent self-governs AND downstream MCPs are gated	Load both servers

Both approaches are complementary and can run side-by-side.

Develop

npm install
npm test       # vitest unit tests
npm run build  # tsup → dist/
npm run typecheck