# toolcallwitness-mcp

A hosted MCP server that records what an agent called: receipts, approval checks, risk summaries and audit exports. Useful when you have to show which tool ran.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI
- Author: clauxel
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/clauxel/tool-call-witness-mcp
- Updated: 2026-05-19
- HTML: https://affpapa.org/en/nevibecode/mcp/toolcallwitness-mcp

## What it does

Four tools ship with it: `create_witness_receipt`, `check_approval_rule`, `summarize_risk_timeline` and `export_audit_receipt`. One writes a receipt for a call, one tests an approval rule, one builds a timeline of risk, one exports a receipt for review. Arguments get redacted along the way and results are condensed into short summaries.

This is a paid remote server, not something you run locally. It speaks Streamable HTTP, so your client has to handle remote servers of that kind. Production traffic needs a bearer token obtained from the product site, passed as `Authorization: Bearer <token>`. Hitting `/mcp` in a browser without credentials returns a JSON error rather than internal detail.

Setup is four steps: pick a plan on the site, get or request an API token, point your MCP client at `https://toolcallwitness.clauxel.com/mcp`, then send JSON-RPC with the token. The repository holds public documentation only — the production source is not there, so you cannot audit the service locally.

## Requirements
- Remote server: `https://toolcallwitness.clauxel.com/mcp`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http toolcallwitness-mcp https://toolcallwitness.clauxel.com/mcp
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add toolcallwitness-mcp --url https://toolcallwitness.clauxel.com/mcp
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "toolcallwitness-mcp": {
            "url": "https://toolcallwitness.clauxel.com/mcp"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "toolcallwitness-mcp": {
            "type": "http",
            "url": "https://toolcallwitness.clauxel.com/mcp"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http toolcallwitness-mcp https://toolcallwitness.clauxel.com/mcp
```

## When to use it
- record which tool an agent invoked and with what arguments
- get evidence that a risky call was approved
- keep sensitive arguments out of the call log
- build a timeline of risk across several calls
- export receipts for review or audit

## Example prompts
- Create a receipt for the last tool call with `create_witness_receipt`
- Run `check_approval_rule` to see whether this call needs approval
- Summarize risk for the session with `summarize_risk_timeline`, then export it via `export_audit_receipt`

## Similar
- [mcpoauthscopegate-mcp](https://affpapa.org/en/nevibecode/mcp/mcpoauthscopegate-mcp.md) — MCP server
- [convex-optimize](https://affpapa.org/en/nevibecode/skills/convex-optimize.md) — skill · 36300 installs
- [dd-audit-ai-activity](https://affpapa.org/en/nevibecode/skills/dd-audit-ai-activity.md) — skill · 433 installs
- [dt-obs-log-semantic-mapping](https://affpapa.org/en/nevibecode/skills/dt-obs-log-semantic-mapping.md) — skill · 1400 installs
- [dd-audit-cost-spike-investigation](https://affpapa.org/en/nevibecode/skills/dd-audit-cost-spike-investigation.md) — skill · 473 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md