agentseal-mcp

MCP server for AgentSeal. Tamper-proof audit logs for AI agents, using SHA-256 hash chains.

Every agent action is recorded in a hash chain. With this, you can actually prove to your clients that your agent did what it said it did.

Setup

Claude Desktop

Add to your claude_desktop_config.json:

config.json

{
  "mcpServers": {
    "agentseal": {
      "command": "npx",
      "args": ["-y", "agentseal-mcp"],
      "env": {
        "AGENTSEAL_API_KEY": "as_sk_your_key_here"
      }
    }
  }
}

Restart Claude Desktop after saving.

Cursor / Other MCP hosts

Same configuration — add the server with your API key.

Environment variables

Variable	Required	Description
`AGENTSEAL_API_KEY`	Yes	Your API key from agentseal.io
`AGENTSEAL_URL`	No	Custom API base URL (defaults to production)

Tools

`record_action`

Record an agent action to the audit trail. Call this after significant actions to create a cryptographically chained record of what happened and why.

Parameter	Type	Required	Description
`agent_id`	string	Yes	Identifier for the agent (e.g. `research-bot`)
`action_type`	string	Yes	What kind of action (e.g. `email:send`, `file:write`, `api:call`)
`action_params`	object	No	Details of the action
`reasoning`	string	No	Why the agent decided to take this action
`authorized_by`	string	No	Who or what approved the action

Returns a sequence number and SHA-256 hash confirming the entry was chained.

`query_actions`

Look up previously recorded actions from the audit trail. Use this to check what actions have been taken or recall past decisions.

Parameter	Type	Required	Description
`agent_id`	string	No	Filter by agent
`action_type`	string	No	Filter by action type
`limit`	number	No	Max entries to return (default 20)

`verify_chain`

Verify the integrity of the hash chain. Each entry's SHA-256 hash includes the previous entry's hash — if any record was modified, the chain breaks and this reports where.

Parameter	Type	Required	Description
`agent_id`	string	No	Verify chain for a specific agent. If omitted, verifies all entries.

Returns the number of entries verified and whether the chain is intact.

How it works

Each recorded action is hashed with SHA-256. That hash includes the previous entry's hash, forming a chain. Modify any record and every hash after it changes — verify_chain catches it instantly.

Get an API key

Python SDK

For direct integration without MCP: pip install agentseal-sdk. See agentseal-sdk.

License

MIT