MCP Server

PeerCat MCP Server

Connect any MCP-compatible LLM or AI application to PeerCat's AI capabilities using the Model Context Protocol.

Important: All Endpoints Disabled by Default

For your protection, all endpoints are disabled by default. LLMs can be unpredictable and might make expensive API calls without your consent. You must explicitly enable each endpoint and model you want to use, and set spending limits before the MCP server will make any requests.

What is MCP?

The Model Context Protocol (MCP) is an open standard that allows AI applications to connect to external tools and data sources. The PeerCat MCP server enables any MCP-compatible LLM or AI agent to access PeerCat's image generation, text AI, and research capabilities.

MCP is supported by a growing ecosystem of AI tools including Claude Desktop, Claude Code, Cursor, Windsurf, and many other LLM-powered applications. Any application that implements the MCP standard can connect to PeerCat's services through this server.

Installation

1. Install the MCP Server

bash
npm install -g @peercat/mcp-server

2. Get your API Key

Create an account at PeerCat and generate an API key from your dashboard.

3. Configure Your MCP Client

Add the PeerCat MCP server to your client's configuration. The exact file location varies by application:

  • • Claude Desktop: claude_desktop_config.json
  • • Claude Code: .claude/settings.json or via /mcp command
  • • Cursor: .cursor/mcp.json
  • • Other clients: Check your application's MCP configuration documentation
MCP Server Configuration
1{
2  "mcpServers": {
3    "peercat": {
4      "command": "npx",
5      "args": ["@peercat/mcp-server"],
6      "env": {
7        "PEERCAT_API_KEY": "your-api-key"
8      },
9      "config": {
10        "endpoints": {
11          "image": {
12            "enabled": false,
13            "models": {
14              "flux-1.1-pro": false,
15              "flux-dev": false,
16              "flux-schnell": false,
17              "sdxl": false,
18              "dall-e-3": false,
19              "imagen-4": false
20            }
21          },
22          "text": {
23            "enabled": false,
24            "models": {
25              "claude-opus-4.5": false,
26              "claude-sonnet-4.5": false,
27              "claude-haiku-4.5": false,
28              "gpt-4o": false,
29              "gpt-4o-mini": false,
30              "gemini-2.5-flash": false
31            }
32          },
33          "research": {
34            "enabled": false
35          }
36        },
37        "spending": {
38          "dailyLimit": 0,
39          "weeklyLimit": 0,
40          "alertAt": 0,
41          "hardStop": true
42        }
43      }
44    }
45  }
46}

4. Enable the endpoints you need

Explicitly enable only the models you want your LLM to use:

Example: Enable specific models with limits
{
  "endpoints": {
    "image": {
      "enabled": true,
      "models": {
        "flux-dev": true,
        "flux-schnell": true,
        "sdxl": true
      }
    },
    "text": {
      "enabled": true,
      "models": {
        "claude-haiku-4.5": true,
        "gpt-4o-mini": true,
        "gemini-2.5-flash": true
      }
    }
  },
  "spending": {
    "dailyLimit": 10,
    "weeklyLimit": 50,
    "alertAt": 40,
    "hardStop": true
  }
}

Configuration Reference

SettingDefaultDescription
endpoints.image.enabledfalseMaster toggle for image generation
endpoints.image.models.*falseIndividual image model toggles
endpoints.text.enabledfalseMaster toggle for text/chat AI
endpoints.text.models.*falseIndividual text model toggles
endpoints.research.enabledfalseToggle for deep research
spending.dailyLimit0Max daily spend in USD (0 = disabled)
spending.weeklyLimit0Max weekly spend in USD
spending.alertAt0Alert when spend reaches this amount
spending.hardStoptrueStop all calls when limit reached

Why All Disabled by Default?

Prevent Unexpected Costs

LLMs can be unpredictable. An AI assistant might decide to generate 100 images or make expensive API calls to Opus 4.5 without your explicit consent. Disabled by default protects you.

Granular Model Control

Enable only the models you need. Keep expensive models like Opus 4.5 disabled while allowing cheaper alternatives like Haiku. Full control over what your LLM can access.

Spending Limits

Set daily and weekly spending limits. When hardStop: true, the MCP server immediately blocks all calls when limits are reached, not just warns.

Explicit Opt-In

You must consciously decide which capabilities to enable. This ensures you understand and accept the potential costs before any LLM can make API calls through PeerCat.