buildshield-mcp 0.1.1

MCP server for BuildShield — AI cost intelligence for agents

buildshield-mcp

MCP server for BuildShield — AI cost intelligence for agents.

Install

pip install buildshield-mcp

Configure (Claude Desktop / Cursor)

Add to your MCP config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "buildshield": {
      "command": "buildshield-mcp",
      "env": {
        "BUILDSHIELD_API_KEY": "bsk_..."
      }
    }
  }
}

Get your API key at thebuildshield.com/settings.

buildshield-mcp is the MCP server command. Keep it inside your MCP client configuration; it waits on stdio and is not meant to print a normal help screen.

Terminal smoke tests

The package also installs CLI wrappers for the same tool handlers:

buildshield_check_cost gpt-4.1 50000 10000
buildshield_get_pricing claude-sonnet-4.6
buildshield_recommend_model "code review" --min-context-tokens 200000

You can also use the grouped command form:

buildshield check-cost gpt-4.1 50000 10000

Tools

Tool	Description	Requires key?
buildshield_check_cost	Exact cost for model + token count	No — offline
buildshield_recommend_model	Best models for a task with filters	No — offline
buildshield_get_pricing	Browse all model pricing	No — offline
buildshield_optimize_prompt	Rule-based prompt compression (20-40% savings)	No — offline
buildshield_check_budget	Check remaining budget for your API key	Yes
buildshield_log_usage	Log usage to your dashboard	Optional

Examples

"What does claude-sonnet-4.6 cost for 50K input + 10K output tokens?"
→ buildshield_check_cost(model="claude-sonnet-4.6", input_tokens=50000, output_tokens=10000)

"What's the cheapest model with at least 200K context?"
→ buildshield_recommend_model(task="long document analysis", min_context_tokens=200000)

"Compress this system prompt"
→ buildshield_optimize_prompt(text="...", aggressiveness="medium")