BIGHUB

@bighub/bighub-mcp

BIGHUB's Better Decision layer as an MCP server. Turn proposed IT agent actions into better decisions before they run from any MCP-compatible client.

MCP server for better decisions on IT agent actions.

MCP client

↓ @bighub/bighub-mcp ↓ BIGHUB API ↓ proposed action → Decision Packet → DecisionBrain → decision signals → run / review / context / block

---

Table of contents

Start here

• Install
• Quickstart
• When to use bighub-mcp
• Typical MCP Loop
• Better Decision primitives
• Trajectory-aware evaluation

Tool reference

• Better Decision primitives · Core loop · Actions · Outcomes · Decision cases · Precedents · Calibration · Multi-signal retrieval · Insights · Simulations · Learning · Features · Runtime ingestion · System integrations · Operating constraints · Approvals & kill switch · Events · Webhooks · API keys · Auth · Utility

Reference

• Environment Variables · Local Development · Links

---

Install

npm install @bighub/bighub-mcp

Requires Node.js 18+.

---

Quickstart

Set your API key:

export BIGHUB_API_KEY=your_api_key

Run the server in stdio mode:

npx @bighub/bighub-mcp

Add it to your MCP client configuration:

{

"mcpServers": { "bighub": { "command": "npx", "args": ["@bighub/bighub-mcp"], "env": { "BIGHUB_API_KEY": "your_api_key" } } } }

Works with MCP-compatible clients such as Claude Desktop and Cursor.

---

When to use bighub-mcp

Use this server when your MCP-connected agent proposes IT actions such as:

• granting Okta access
• rotating credentials
• deploying or rolling back services
• changing CI/CD, cloud, or IAM settings
• exporting sensitive data
• posting incident or support updates
• resolving actions that need human review

If your agent only answers questions or performs read-only lookups, you don't need this. It's designed for agents that make decisions with real-world consequences.

---

Typical MCP Loop

1. Send the proposed IT action to bighub_decide
2. Let BIGHUB build or use a Decision Packet
3. Run DecisionBrain and return execution mode, review/context flags, and a decision path when available
4. Execute only when can_run is true and no blocking flags are set
5. Route to review when BIGHUB asks for human approval or modification
6. Optionally report the outcome later

bighub_decide

→ bighub_build_packet / bighub_run_brain → bighub_resolve_review when needed → bighub_report_outcome later, optional

---

Better Decision primitives

The modern MCP surface is centered on the decision flow:

bighub_decide and bighub_run_brain return a normalized decision object first: recommended_action, mode, can_run, needs_review, needs_more_context, should_not_run, risk, confidence, expected_regret, reason, system, selected_model when available, decision_path when available, and counts for verification steps and obligations. The original backend payload is still available under raw for advanced callers. Legacy tools such as bighub_actions_evaluate remain available for compatibility.

bighub_build_packet uses the same canonical JSON SHA-256 hashing strategy as the Python SDK, so packet hashes match across SDK and MCP clients.

---

Trajectory-aware evaluation

BIGHUB evaluates actions not only in isolation, but also in the context of what happened before. As deployment, review, and outcome history accumulates, similar IT actions can receive better recommendations over time.

---

Complete Tool Reference

BIGHUB tools are organized by domain. The core loop tools are listed first.

---

Core loop

---

Actions

---

Outcomes

---

Decision cases

---

Precedents

---

Calibration

---

Multi-signal retrieval

---

Insights

---

Simulations

---

Learning

---

Features

---

Runtime ingestion

---

System integrations

Supported system providers include okta, slack, github, sentry, datadog, aws_cloudtrail, terraform, kubernetes,