VeriSwarm MCP Server

The VeriSwarm MCP Server exposes the VeriSwarm platform to AI agents through the Model Context Protocol. Any MCP-compatible client — Claude Desktop, Cursor, or custom runtimes — can check trust scores, report events, manage agents, and interact with Guard, Passport, and Vault.

Installation

pip install veriswarm-mcp

Requires Python 3.10+.

Authentication

The server authenticates using environment variables:

Variable	Required	Description
VERISWARM_API_URL	No	API base URL (default: https://api.veriswarm.ai)
VERISWARM_API_KEY	Yes (or AGENT_KEY)	Platform API key for workspace operations
VERISWARM_AGENT_KEY	Yes (or API_KEY)	Agent-scoped key for self-reporting and credentials

For most setups, you only need VERISWARM_API_KEY. Set VERISWARM_AGENT_KEY when the server runs on behalf of a specific agent.

Available Tools (26)

Trust Scoring (5 tools)

check_trust, check_decision, get_my_score, get_score_history, explain_score

Event Reporting (4 tools)

report_action, report_tool_call, report_interaction, report_incident

Guard Security (7 tools)

scan_tool, check_tool_allowed, get_findings, kill_agent, unkill_agent, tokenize_pii, rehydrate_pii

Passport Identity (4 tools)

get_credentials, verify_credential, verify_identity, check_delegation

Vault Audit (3 tools)

query_ledger, verify_chain, export_evidence

Agent Management (3 tools)

register_agent, get_agent, list_agents

Platform (1 tool)

platform_status

For detailed parameter descriptions, see the Python SDK or Node.js SDK documentation.

Client Configuration

Add the following to your MCP client configuration:

{
  "mcpServers": {
    "veriswarm": {
      "command": "python",
      "args": ["-m", "src"],
      "env": {
        "VERISWARM_API_URL": "https://api.veriswarm.ai",
        "VERISWARM_API_KEY": "vs_your_platform_key"
      }
    }
  }
}

Config file locations:

• Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
• Claude Desktop (Windows): %APPDATA%\Claude\claude_desktop_config.json
• Cursor: .cursor/mcp.json in your project, or Cursor Settings > MCP

If using a virtual environment, replace "python" with the full path to that environment's Python binary.

Usage Examples

Check trust and make decisions

"What is the current trust score for agent agt_abc123?"

"Can agent agt_abc123 perform a file_write action on production_db?"

Report events

"Report that agent agt_abc123 successfully called search_documents in 230ms."

"Report a high-severity prompt injection attempt from agent agt_xyz789."

Passport credentials

"Issue me a Passport credential."

"Verify this Passport credential: eyJhbGciOi..."

Guard and Vault

"Kill-switch agent agt_abc123 — suspected data exfiltration."

"Tokenize any PII in this text before sending it to the external tool."

"Rehydrate the tokenized response so I can read the original values."

"Verify the integrity of the last 200 Vault ledger entries."

Guard Proxy Mode

Agents can connect through a Guard Proxy MCP server instead of connecting directly to upstream tool servers. The proxy transparently intercepts every tool call to apply PII tokenization, policy enforcement, injection scanning, and audit logging before forwarding requests. Configure the proxy endpoint in your MCP client config just like any other MCP server.

Rate Limits

The MCP server inherits rate limits from the VeriSwarm REST API. Limits vary by plan tier — see your workspace billing settings for details. Rate-limited requests return HTTP 429.