# webaesbyamin/agent-receipts [Health: Active]

**Category:** 🛠️ Other Tools and Integrations  
**Repository:** https://github.com/webaesbyamin/agent-receipts  
**GitHub Stars:** 1  
**npm Downloads (last month):** 276  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/webaesbyamin-agent-receipts

## Description
Cryptographic accountability for AI agents. Ed25519-signed receipts for every MCP tool call — constraints, chains, AI judgment, invoicing, local dashboard.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

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

## Documentation

## What webaesbyamin/agent-receipts MCP server does

webaesbyamin/agent-receipts provides an MCP interface for two related jobs: maintaining structured memory across agent sessions and producing verifiable receipts for memory events and agent actions. Memory is organized as entities, observations, relationships, preferences, and facts rather than a simple key-value list. Each stored observation can be traced to the conversation that created it.

The webaesbyamin/agent-receipts MCP server also records signed JSON receipts. A receipt contains an Ed25519 signature, hashes for the input and output, timestamps, an agent identifier, and links in a provenance chain. This lets a verifier check the record without relying on the agent's own mutable log or retaining the original raw payload in the receipt.

## How it works

The server runs locally through the `@agent-receipts/mcp-server` npm package. Its default storage is a SQLite database under `~/.agent-receipts/`, with FTS5 used for full-text memory search.

A typical session starts with `memory_context`, which loads relevant known information. The agent can call `memory_observe` when it learns a durable fact, `memory_recall` to search existing memory, and `memory_forget` to soft-delete information while preserving an audit trail. Significant work can be recorded with `track_action`. The system prompt supplied by the project tells an agent when to use these operations, so memory and action tracking do not depend only on manual calls.

Receipts can be checked with verification tools, and memory can be exported as a portable bundle. An export includes entities, observations, relationships, their creating receipts, and the public key needed for verification.

## Setup and configuration

Install the MCP server with `npx` and add it to an MCP client using the command `npx @agent-receipts/mcp-server`. Claude Code uses `claude mcp add agent-receipts -- npx @agent-receipts/mcp-server`. Claude Desktop and Cursor can use an MCP configuration entry with `command` set to `npx` and `args` set to `[@agent-receipts/mcp-server]`.

After adding the server, generate the client-specific system prompt with `npx @agent-receipts/cli prompts claude-code`. The generated text should be copied into project instructions or the agent's system prompt. No required environment variables or external credentials are documented.

The optional dashboard starts with `npx @agent-receipts/dashboard` and is available at `localhost:3274`. CLI commands include statistics, memory inspection, and memory export, such as `npx @agent-receipts/cli stats` and `npx @agent-receipts/cli memory export`.

## Tools and capabilities

- Persistent memory context, observation, recall, and forgetting.
- Action tracking and receipt lifecycle operations.
- Ed25519 receipt verification and public-key access.
- Constraints, AI judgments, invoicing, and bundle operations.
- Local receipt, memory, chain, agent, and related dashboard views.

The underlying SDK includes methods for starting and completing pending receipts, retrieving or listing receipts, checking judgments, and cleanup. The CLI also provides memory subcommands and terminal summaries.

## Limitations and notes

webaesbyamin/agent-receipts is local-first, so its documented storage model is a SQLite database on the host running the server. The system prompt is an important part of the workflow: without it, an agent may not automatically call memory or action tools at the intended points.

Receipts hash inputs and outputs rather than storing the raw data in the receipt itself. Verification establishes the integrity of the signed record and its chain; it does not by itself establish that an agent's underlying action was appropriate or that an asserted result is factually correct. Exported bundles are intended to be independently verifiable using the included public key.

_Full upstream README: https://allmcps.com/mcp/webaesbyamin-agent-receipts/readme_