by Kaval
Kaval MCP gives AI agents access to Kaval's fact checks, signed receipts and watched sources. An agent can ask whether an action still rests on current facts before it acts.
PluginContact for pricing
Step 1: Get a Kaval API key from the documentation quickstart.
Step 2: Add an mcpServers entry named kaval to your MCP client config, with command npx and args -y @usekaval/mcp.
Step 3: Set the KAVAL_API_KEY environment variable in that entry.
Step 4: Create a publisher, then ask the agent to add a source such as a policy URL using add_source.
Step 5: Before a consequential step, have the agent call check with the action and a materiality level, and act on the ALLOW, REVIEW or BLOCK answer.
It registers 29 tools. They cover checks, receipts, watched sources, extraction schemas and runs, extraction packages and fact imports.
A decision of ALLOW, REVIEW or BLOCK, the reason codes, the facts it ran on and a signed receipt. The receipt names the rule version so the decision can be re-derived.
No. The pricing page says Kaval is not a scrubber and sits under an existing rules engine, keeping the published facts current and evidence-anchored.
The server runs over stdio and needs one environment variable, KAVAL_API_KEY.
Requirements
Vaara Server MCPVaara checks each action an AI agent or other autonomous software takes against your rules before it runs, and writes a signed receipt anyone can verify later. It is open source and runs on your own machines.
Verax MCPVerax is an open-source MCP server that sits between an agent and its tools. It checks each tool call against a policy and writes a signed allow, deny or defer record before the call runs.
Arbiter MCPDecision arbitration, action routing, workflow control, incident response, and memory.