@agent-receipts/mcp-server

Connect any MCP-compatible AI agent to cryptographic accountability in 3 lines of config. Every action gets a signed receipt. Every memory is provable. No API key, no account, no server — everything runs locally.

Try the Interactive Demo · Full Documentation

Quick Start

Add to your MCP client config and every agent action gets a cryptographic receipt automatically.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "agent-receipts": {
      "command": "npx",
      "args": ["@agent-receipts/mcp-server"]
    }
  }
}

Claude Code

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "agent-receipts": {
      "command": "npx",
      "args": ["@agent-receipts/mcp-server"]
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "agent-receipts": {
      "command": "npx",
      "args": ["@agent-receipts/mcp-server"]
    }
  }
}

Tools

The MCP server exposes 24 tools:

Tool	Description	Key Parameters
track_action	Track an agent action with automatic hashing	action, input, output, constraints
create_receipt	Create a receipt with pre-computed hashes	action, input_hash, output_hash, constraints
complete_receipt	Complete a pending receipt with results	receipt_id, output, status
verify_receipt	Verify the cryptographic signature of a receipt	receipt_id
get_receipt	Retrieve a receipt by ID	receipt_id
list_receipts	List receipts with optional filtering	agent_id, status, chain_id
get_chain	Get all receipts in a chain ordered by timestamp	chain_id
get_public_key	Export the Ed25519 public key for verification	—
judge_receipt	Start AI Judge evaluation of a receipt	receipt_id, rubric
complete_judgment	Complete a pending judgment with results	receipt_id, verdict, score, criteria
get_judgments	Get all judgments for a receipt	receipt_id
cleanup	Delete expired receipts (TTL)	dry_run
generate_invoice	Generate an invoice from receipts	from, to, format, agent_id
get_started	Show a getting-started guide with usage examples	--
memory_observe	Store a memory observation about an entity	entity_name, entity_type, content, ttl_seconds
memory_recall	Search and retrieve stored memories	query, entity_type, scope
memory_forget	Soft-delete an observation or entity	entity_id or observation_id
memory_entities	List known entities with filtering	entity_type, scope, query
memory_relate	Create a relationship between two entities	from_entity_id, to_entity_id, relationship_type
memory_provenance	Get the provenance chain for an observation	observation_id
memory_context	Get a structured memory context summary	entity_type, scope, agent_id
memory_audit	Generate a memory operations audit report	agent_id, from, to
memory_export_bundle	Export memories as a portable, verifiable bundle	entity_ids, include_receipts
memory_import_bundle	Import and verify a memory bundle	bundle, skip_existing

Configuration

Environment Variable	Description	Default
AGENT_RECEIPTS_DATA_DIR	Data directory path	~/.agent-receipts
AGENT_RECEIPTS_AGENT_ID	Default agent ID	default-agent
AGENT_RECEIPTS_ORG_ID	Organization ID	default-org
AGENT_RECEIPTS_ENVIRONMENT	Environment label	production
RECEIPT_SIGNING_PRIVATE_KEY	Ed25519 private key (hex)	Auto-generated
RECEIPT_SIGNING_PUBLIC_KEY	Ed25519 public key (hex)	Derived from private

Programmatic Usage

The engine and storage classes are exported for use by @agent-receipts/sdk and custom integrations:

import { ReceiptEngine, ReceiptStore, KeyManager, ConfigManager } from '@agent-receipts/mcp-server'

const store = new ReceiptStore('/path/to/data')
await store.init()
const keyManager = new KeyManager('/path/to/data')
await keyManager.init()
const configManager = new ConfigManager('/path/to/data')
await configManager.init()
const engine = new ReceiptEngine(store, keyManager, configManager)

const receipt = await engine.track({ action: 'my_action', input: 'data' })

License

MIT