Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@haver-governance-mcp Check permission for intake assistant to read labs"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Identity, permission, audit, and recall for healthcare AI agents, delivered over the Model Context Protocol (MCP).
As organizations deploy more AI agents into clinical and operational workflows, the missing layer is governance: knowing which agent is acting, whether it is allowed to, having a tamper-evident record of what it did, and being able to revoke it. Haver Governance MCP gives any MCP-compatible client (such as Claude) a small set of tools to enforce and record that governance at runtime.
This is an open-source reference implementation from Haver AI, the governance layer for healthcare AI.
Tool | Pillar | What it does |
| Identity | Registers an agent with a stable id, owner, role, and permission scopes. |
| Permission | Decides whether an agent may take an action on a resource. Denies unknown or recalled agents. Logs the decision. |
| Audit | Appends a tamper-evident, hash-chained record of an agent action. |
| Audit | Returns recent audit entries, filterable by agent or resource. |
| Audit | Recomputes the hash chain and reports whether the log was tampered with. |
| Recall | Revokes an agent so every future permission check denies it. |
| Safety | Heuristically flags and redacts common PHI (SSN, MRN, email, phone, dates). |
Scopes use a simple action:resource form, for example read:labs or write:notes, and support * wildcards (read:*, *:*).
Related MCP server: Agent Audit Trail MCP Server
Run it directly with npx (no global install needed):
npx -y haver-governance-mcpAdd this to your claude_desktop_config.json:
{
"mcpServers": {
"haver-governance": {
"command": "npx",
"args": ["-y", "haver-governance-mcp"]
}
}
}Restart Claude, and the seven governance tools appear.
register_agent an intake assistant with scopes: ["read:labs"].
check_permission for read on labs returns allow; write on notes returns deny.
record_action logs each real action it takes, flagging contains_phi when relevant.
If the agent misbehaves, recall_agent revokes it; future checks deny.
verify_audit_integrity proves the log has not been altered.
All state is stored locally as plain files: agent records in agents.json and the audit log in audit.log, under ~/.haver-governance by default (override with the HAVER_GOV_DIR environment variable). Nothing is transmitted anywhere. The audit log is append-only and hash-chained for tamper evidence. See PRIVACY.md.
The scan_phi tool is a heuristic guardrail to catch obvious leaks. It is not a certified de-identification tool and should not be relied on as one.
npm install
npm run build
node dist/index.js # runs on stdioRequires Node.js 18 or later. Built with the official @modelcontextprotocol/sdk and Zod.
This package is published to npm and listed in the official MCP Registry. See PUBLISHING.md for the exact steps.
Haver AI is the governance layer for healthcare AI: identity, permission, audit, and recall for every agent your organization deploys. Learn more at haver.ai.
MIT. See LICENSE.
Decide whether a registered agent may perform an action on a resource, based on its scopes and status. Recalled or unknown agents are always denied. The decision is written to the audit log.
Inputs: action (required), agent_id (required), resource (required).
Return recent audit entries, optionally filtered by agent or resource.
Inputs: limit, agent_id, resource.
Revoke an agent so all future permission checks deny it. Use when an agent misbehaves or is decommissioned. The recall is audited.
Inputs: reason (required), agent_id (required).
Append a tamper-evident audit record for something an agent did. Entries are hash-chained, so any later modification is detectable.
Inputs: action (required), outcome (required), agent_id (required), resource (required), +2 more.
Register an AI agent with a stable identity, owner, role, and permission scopes. Scopes use the form 'action:resource' (e.g. 'read:labs'), and '*' wildcards are allowed.
Inputs: name (required), role (required), owner (required), scopes (required), +1 more.
Heuristically scan text for common protected health information (SSN, MRN, email, phone, dates) and return findings plus a redacted copy. A guardrail, not a certified de-identification tool.
Inputs: text (required).
Recompute the audit hash chain and report whether the log is intact or has been tampered with.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 7 tools
Each tool has a unique purpose: permission checking, audit trail retrieval, agent recall, action recording, agent registration, PHI scanning, and integrity verification. No two tools overlap or could be confused.
All tools follow a consistent verb_noun pattern (e.g., check_permission, register_agent, scan_phi) with imperative verbs and descriptive nouns. No mixing of styles or vague names.
Seven tools is well-scoped for a governance server covering agent lifecycle, permissions, audit, and integrity. Each tool earns its place without being excessive or insufficient.
The tool surface covers core CRUD-like operations for agents (register, recall), permissions (check), audit (record, get, verify integrity), and adds a guardrail (scan_phi). No obvious gaps for the domain.
Governance for AI agents in health systems: inventory, scoped permissions, evidence, revocation.
Context and control for any AI. Agent identity, scoped access, and proof across tools.
Register every AI agent, log every action, prove it. EU AI Act compliance built in.
Identity, authorization, audit trails, and revocable permissions for AI agents accessing MCP tools.
