@achatainga/burnguard-mcp
v1.0.1
Published
Pre-Execution Token Budget Firewall & Cost Telemetry Sentinel for AI Coding Agents
Maintainers
Readme
BurnGuard MCP
Pre-Execution Token Budget Firewall & Cost Telemetry Sentinel for AI Coding Agents.
The Problem
When autonomous agents run without financial guardrails:
- Accidental Giant File Reads: Ingesting lockfiles, minified bundles, or binary dumps consuming 100k+ tokens in a single second.
- Infinite Debugging Token Bleed: Spending $20+ across repetitive failed test iterations without user notification.
- Surprise Invoices: Developers discovering massive bill spikes on their Anthropic, OpenAI, or Google Cloud bills.
The Solution
BurnGuard MCP enforces strict financial and context boundaries:
- Pre-Execution Cost Estimator: Calculates exact token burn and financial cost across frontier models prior to execution.
- Bloat Path Interceptor: Hard-blocks reading or dumping giant build directories (
node_modules,dist,vendor,.git). - Hard Session Budget Cap: Halts agent execution with exit code 2 when session spend crosses developer-set limits ($2.00, $5.00, etc.).
Available MCP Tools
estimate_token_cost
Calculates character counts, token equivalents, and dollar costs across frontier models.
inspect_payload_bloat
Identifies dangerous context-bloating file paths and returns targeted mitigation advice.
enforce_budget_cap
Audits cumulative spend against hard session limits and halts runaway processes.
Live Verification
node tests/run_live_test.mjsLicense
MIT © 2026 BurnGuard Systems
