Figuard Mcp
Give Claude Code or Cursor a pre-flight gate so agent tool calls cannot blow past a FiGuard budget without an auditable approve-or-deny decision.
Overview
FiGuard MCP is a MCP server for the Operate phase that pre-authorizes AI agent spend against budgets and audits each decision before limits are exceeded.
What is this MCP server?
- Pre-flight spend authorization before agent actions execute
- Set budgets and enforce hard limits on agent-related spend
- Audit trail for every authorization decision
- stdio npm package figuard-mcp v1.0.0 for MCP clients
- Optional FIGUARD_API_KEY with public sandbox when unset
- Server version 1.0.0 per MCP registry schema
- Single npm package figuard-mcp with stdio transport
- Two optional env vars: FIGUARD_API_KEY (secret) and FIGUARD_BASE_URL
Community signal: 1 GitHub stars.
What problem does it solve?
Autonomous agents can rack up API bills in minutes with no single place to cap spend or prove what was approved.
Who is it for?
Solo builders running production agent workflows on metered LLM or SaaS APIs who need a simple budget gate without building custom billing middleware.
Skip if: Teams that only run local, zero-cost dev prompts with no external paid tools, or anyone who needs full accounting/invoicing rather than agent pre-flight checks.
What do I get? / Deliverables
Your agent asks FiGuard for spend authorization first, stays inside budget, and leaves an audit log you can review after each session.
- Budget-enforced pre-flight checks on agent spend actions
- Per-decision audit records for agent authorization
- Running FiGuard MCP bridge via stdio in your agent config
Recommended MCP Servers
Journey fit
Solo builders feel agent spend pain after shipping automations; spend authorization belongs on the operate shelf where production agents run continuously against paid APIs. Monitoring covers cost guardrails and decision logs—not building features, but watching whether each agent action stays inside policy.
How it compares
MCP spend-authorization guardrail, not an agent skill or generic API wrapper without policy enforcement.
Common Questions / FAQ
Who is io.github.figuard/figuard-mcp for?
Indie and solo builders who ship AI agents in Claude Code, Cursor, or similar clients and want enforced budgets plus audited spend decisions before tools run.
When should I use io.github.figuard/figuard-mcp?
Use it when agent loops call paid APIs in production or staging and you need pre-flight approval instead of discovering overages on the monthly bill.
How do I add io.github.figuard/figuard-mcp to my agent?
Add the npm figuard-mcp stdio server to your MCP client config, set FIGUARD_API_KEY and FIGUARD_BASE_URL if you have an account, or omit them to try the public sandbox.