Skip to main content
Glama
@vaibot/mcp-server

@vaibot/mcp-server

Official
by vaibot-io

@vaibot/mcp-server

npm MCP Registry

VAIBot governance circuit-breaker as an MCP server.

Exposes 4 tools to any MCP-compatible AI agent (Claude Code, Codex, ChatGPT, OpenClaw):

Tool

Description

vaibot_decide

Pre-execution risk + policy decision. Call before any risky action.

vaibot_finalize

Report actual outcome after execution. Closes the governance receipt.

vaibot_receipts

List recent governance receipts with optional filters.

vaibot_approve

Approve or deny a pending action from the dashboard or agent.

Quick start

VAIBOT_API_KEY=vb_live_xxx \
VAIBOT_API_BASE_URL=https://api.vaibot.io \
npx @vaibot/mcp-server

Related MCP server: governed-mcp

Claude Desktop config

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "vaibot": {
      "command": "npx",
      "args": ["-y", "@vaibot/mcp-server"],
      "env": {
        "VAIBOT_API_KEY": "vb_live_xxx",
        "VAIBOT_API_BASE_URL": "https://api.vaibot.io"
      }
    }
  }
}

Claude Code / Codex (stdio)

# .mcp.json in your project root
{
  "mcpServers": {
    "vaibot": {
      "command": "npx",
      "args": ["-y", "@vaibot/mcp-server"],
      "env": {
        "VAIBOT_API_KEY": "vb_live_xxx"
      }
    }
  }
}

OpenClaw gateway plugin (coming in Phase 6 final)

The vaibot-guard-bridge plugin will point guardBaseUrl at the MCP server endpoint instead of localhost:39111. Config change only — no plugin code changes needed.

Remote URL transport (HTTP)

If your agent supports MCP over HTTP (Remote URL mode), connect directly to the VAIBot API without installing anything:

URL:   https://api.vaibot.io/v2/mcp
Token: Bearer <your-api-key>

The HTTP endpoint speaks JSON-RPC 2.0 and supports all 4 tools. Auth is the same API key as the stdio transport.


Environment variables

Variable

Required

Default

Description

VAIBOT_API_KEY

✅

—

VAIBot API key (vb_stg_xxx or vb_live_xxx)

VAIBOT_API_BASE_URL

—

https://api.vaibot.io

API base URL

Governance flow

Agent wants to run: curl -X POST https://deploy.example.com/release

1. Agent calls vaibot_decide:
   → VAIBot: APPROVAL_REQUIRED (high risk — outbound network call)
   → Returns: run_id, content_hash

2. Human reviews in dashboard, clicks Approve

3. Agent calls vaibot_approve (or dashboard fires callback):
   → VAIBot: ✅ APPROVED

4. Agent executes the action

5. Agent calls vaibot_finalize:
   → VAIBot: receipt updated, outcome=allowed

Available Tools

4 tools
vaibot_approveA

Approve or deny a pending governance action. Use content_hash from vaibot_decide or vaibot_receipts. action="approve" allows the blocked action to proceed; action="deny" permanently blocks it.

Inputs: action (required), content_hash (required).

vaibot_decideA

Ask VAIBot whether an agent action should be allowed, requires approval, or is denied. Call this BEFORE executing any tool that could be risky (exec, file writes, network calls). Returns a run_id to use with vaibot_finalize after execution.

Inputs: tool (required), agent_id (required), session_id (required), cwd, +4 more.

vaibot_finalizeA

Report the actual outcome of an action back to VAIBot after execution. Always call this after a vaibot_decide allow decision to close the governance receipt.

Inputs: run_id (required), outcome (required), error, exit_code, +1 more.

vaibot_receiptsA

List recent governance receipts for the current agent. Use to review what actions were allowed, denied, or are pending approval.

Inputs: tool, limit, decision, risk_level, +1 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.3/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool has a distinct role in the governance workflow: decide pre-checks actions, finalize reports outcomes, receipts lists history, and approve handles pending decisions. There is no overlap or ambiguity between them.

    Naming Consistency4/5

    All tools share the 'vaibot_' prefix and mostly use verb-based naming (decide, finalize, approve). 'receipts' is a noun rather than a verb, deviating slightly from the action-oriented pattern, but the prefix and clear purpose keep it readable.

    Tool Count5/5

    Four tools provide a well-scoped governance surface: pre-check, post-report, list, and approve. This is neither too thin nor excessive for the stated purpose.

    Completeness5/5

    The workflow is fully covered: initiate with decide, execute, finalize with the run_id, review via receipts, and handle pending approvals with approve. There are no obvious missing operations for the governance domain.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that enforces runtime governance on AI agent actions — file access, command execution, delegation chains, and permission escalation.
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      MCP server that enforces governance on agentic decisions with auditable evidence records, providing tools for understanding, calibrating confidence, and navigating handoffs based on policy.
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      Runtime permission, approval, and audit governance for AI agent tool execution, enabling human oversight of risky actions via an MCP server.
      1
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      MCP server that provides human-in-the-loop approval for risky AI agent actions, with durable state and audit logs.
      MIT