io.github.Stratogenic-AI/catalyst

Official

by Stratogenic-AI

Catalyst Governance

Governance middleware for AI agents: permission gates, approvals, compliance scanning, audit ledger.

Permission gates, approvals, compliance scanning, and audit ledger for AI agents.

Topics

ai-governanceauditcompliancellmmcpmcp-serverai-agentsanthropicapproval-workflowaudit-trailclaude-codeeu-ai-actfastmcpguardrailshuman-in-the-looppermission-gatestreamable-httpgovernance-middleware

README.md5 KBView on GitHub

Catalyst MCP Server

Governance middleware for AI agents — hosted MCP server with both SSE and Streamable HTTP transports.

What it does

Catalyst puts a governance layer between your AI agent and consequential actions. Before your agent writes files, sends messages, deploys code, or calls external APIs, it checks in with Catalyst. Depending on your governance mode, actions are allowed, routed to a human for approval, or denied.

Core features:

Pre-execution permission gates with a progressive trust model: observe → advisory → proposal → strict
Human-in-the-loop approval flows — agents submit proposals, humans approve/decline
Compliance scanning across 8 frameworks: GDPR, SOC 2, ISO 27001, HIPAA, PCI-DSS, EU AI Act, Bribery Act, AML/KYC
Hash-linked audit ledger — every agent action is recorded with touched_by.source=mcp
Webhook notifications for proposal.accepted / proposal.declined events

Use with Claude Code

Claude Code is a primary use case Catalyst was designed for. Claude Code governs itself session-locally — it asks before running dangerous commands, but those decisions leave no audit trail, survive no context reset, and sit outside your organisation's approval workflows.

Catalyst externalises that governance. Once connected, every consequential action Claude Code attempts is routed through your policy, logged in an immutable ledger, and (if required) held for human approval in the Catalyst dashboard before Claude Code proceeds.

Connect in one command:

bash

claude mcp add catalyst \
  --transport sse \
  --url https://catalyst.stratogenic.ai/mcp/sse \
  --header "X-API-Key: SGC_your_key_here"

Or with Streamable HTTP (Claude Code 0.10+):

bash

claude mcp add catalyst \
  --transport http \
  --url https://catalyst.stratogenic.ai/mcp \
  --header "X-API-Key: SGC_your_key_here"

Claude Code will automatically call catalyst_check_action before destructive operations and catalyst_await_approval when a proposal is required. Every action is ledgered with touched_by.source=mcp — you get a permanent, verifiable record of what your AI coding agent did and when.

What governance looks like in practice:

Claude Code action	Catalyst response (proposal mode)
`git push --force`	`proposal_required` — held for approval
`rm -rf ./dist`	`allow` — whitelisted build action
Deploy to production	`proposal_required` — held for approval
Edit source files	`allow` — within declared capabilities
Call external API	`proposal_required` — undeclared capability

Set your governance mode to observe to start with a full audit trail only, advisory to flag deviations without blocking, proposal to require human sign-off on sensitive actions, or strict to allow only explicitly whitelisted capabilities.

Get your API key at catalyst.stratogenic.ai.

Connect

Streamable HTTP (preferred — required by Glama, Smithery, and newer clients):

json

{
  "mcpServers": {
    "catalyst": {
      "url": "https://catalyst.stratogenic.ai/mcp",
      "transport": "http",
      "headers": { "X-API-Key": "<your-key>" }
    }
  }
}

SSE (for clients that don't yet support Streamable HTTP):

json

{
  "mcpServers": {
    "catalyst": {
      "url": "https://catalyst.stratogenic.ai/mcp/sse",
      "transport": "sse",
      "headers": { "X-API-Key": "<your-key>" }
    }
  }
}

Get your API key at catalyst.stratogenic.ai.

Tools

Tool	Description
`catalyst_my_governance`	Show your governance profile and plan capabilities
`catalyst_check_action`	Check if a capability is permitted before acting
`catalyst_log_task`	Record a task or action into the governance pipeline
`catalyst_pending_approvals`	List proposals awaiting human sign-off
`catalyst_await_approval`	Poll a specific proposal's approval status
`catalyst_register_approval_webhook`	Register a callback URL for approval events
`catalyst_run_compliance_scan`	Run an automated compliance scan (requires add-on)
`catalyst_list_compliance_frameworks`	List available compliance frameworks
`catalyst_get_dashboard`	Get the governance dashboard snapshot
`catalyst_export_ledger`	Export the hash-linked audit chain

Governance loop

code

catalyst_my_governance()            # understand your mode + capabilities
    ↓
catalyst_check_action("write_file") # allow | proposal_required | deny
    ↓ if proposal_required
catalyst_log_task("Deploy to prod") # creates proposal, returns proposal_id
    ↓
catalyst_await_approval(proposal_id) # poll until approved: true
    ↓ proceed

Install

claude_desktop_config.json

{
  "mcpServers": {
    "catalyst": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://catalyst.stratogenic.ai/mcp"
      ]
    }
  }
}

Desktop config is stdio-only; this bridges via mcp-remote. Native remote: Settings > Connectors.