# pepe

> pepe is an open-source MCP server for developer tools. Load-test a URL, ping it by phase and ramp load from an agent, within set guardrails. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, macOS, Linux, Windows, self-hosted
- **Website:** https://pepe.mhaimdat.com/docs/agents.html
- **Repository:** https://github.com/omarmhaimdat/pepe
- **MCP Registry name:** io.github.omarmhaimdat/pepe
- **Categories:** developer tools, cloud and infrastructure
- **Listing updated:** 2026-11-15

## About

pepe is a Rust HTTP load generator with a live dashboard, and its MCP mode turns the same engine into tools an agent can call. Each call returns pepe's versioned JSON report, so an agent can run a test, ping a target phase by phase (DNS, connect, TLS, first byte, download), ramp load to find where a target stops keeping up, and compare runs.

The safety design is the point of the MCP page. Whoever starts the server sets allowed hosts, a request cap, a rate cap and a concurrency cap, checked before anything is sent; every redirect hop is checked and a tool call cannot loosen them. A dry run reports what would be sent with secrets masked. Exit codes distinguish a usage error or guardrail refusal from a crossed limit.

The README claims one thread sends 160k requests a second on an Apple M4 Pro, and states it is lighter on CPU and memory than wrk and oha.

## Features

- **Host allow-list:** Targets and redirect hops must match an allowed host.
- **Request, rate and concurrency caps:** Limits set at server start that tool calls cannot raise.
- **Dry run:** Shows what would be sent with secrets masked, sending nothing.
- **Phase-split pings:** Breaks each ping into DNS, connect, TLS, first byte and download.
- **Capacity ramp:** Raises load to find where a target stops keeping up.
- **Versioned JSON reports:** Published schema with stable field names.

## Good for

- Let an agent load-test a staging API within strict limits
- Find the capacity of a service by ramping load
- Compare two runs after a change
- Diagnose where time goes in a slow request

## Install

### Claude Code (Terminal)

```sh
claude mcp add pepe -- uvx pepe-http
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "pepe": {
      "command": "uvx",
      "args": [
        "pepe-http"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "pepe": {
      "command": "uvx",
      "args": [
        "pepe-http"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "pepe": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "pepe-http"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "pepe": {
      "command": "uvx",
      "args": [
        "pepe-http"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "pepe": {
      "command": "uvx",
      "args": [
        "pepe-http"
      ]
    }
  }
}
```

## Tools

- `run`: Runs a load test.
- `ping`: Pings a target with phase timings.
- `ramp`: Ramps load to find capacity.
- `api`: Load-tests endpoints from an OpenAPI spec.
- `flow`: Runs a flow of dependent requests.
- `replay`: Replays an access log.
- `logs`: Reads request logs.
- `compare`: Compares runs.

## FAQ

### Can an agent override the guardrails?

No. Flags given at server start hold for every call and a tool call cannot loosen them.

### How do I install the MCP server?

Run uvx pepe-http mcp or the ghcr.io/omarmhaimdat/pepe Docker image with the mcp argument.

### What do the exit codes mean?

0 as asked, 1 could not start or nothing answered, 2 usage error or guardrail refusal, 4 a limit crossed.

### Is it safe to point at any site?

The docs say to load only what you are allowed to load; guardrails make that a setting.

### Is pepe free?

Yes. pepe is open source, so you can use and self-host it for free.

### How do I add pepe to Claude Code?

Run this command in your terminal: claude mcp add pepe -- uvx pepe-http

### What are the best alternatives to pepe?

The closest alternatives to pepe are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/pepe
