"Actual Budget" matching MCP servers:

Screen-reader navigation cost analyzer. Measures how hard it is for assistive-technology users to discover, reach, and operate interactive targets on the web.

Enables coding agents to see the UI they just changed without flooding their context window, returning a structured text summary (console errors, failed requests, overflow, layout shifts, accessibility violations) plus small crops of only the regions that changed since the last capture. All output is fitted into a chosen token budget, with captures stored on disk and evicted automatically.

An MCP server that deterministically scores landing pages on conversion-critical signals and blocks ad spend on low-quality pages to prevent wasted budget.

Validates AI-generated code against actual codebases to catch hallucinations, dead code, and API mismatches before runtime.

Detects and reports when AI coding agents modify files outside their declared task scope, comparing declared globs and categories with actual changes.

Enables AI agents to drive real browser automation through the full upstream Playwright MCP tool surface, powered by a hardened CloakBrowser Chromium runtime, so users can ask for web research, daily automation, or testing in plain English and inspect the actual results. Ships as npm and Docker packages over stdio or Streamable HTTP, with persistent profiles, extension loading, proxy matching, and humanized mouse, keyboard, and scroll behavior.

Enables coding agents to start and operate live kernels on local machines, Modal GPUs, or Colab VMs, running code, notebook cells, shell commands, and file transfers while inspecting variables, plots, and errors. It also runs notebooks, scripts, and web UIs end to end as budget-capped plans that verify expectations and report what failed and what changed.

Enables transforming raw bug notes into structured defect reports (Summary, Steps, Expected vs Actual, Priority) via a prompt template.

tsugime is an MCP server that declares expected correspondences between tools (like index files and actual files) and checks them. It reports any mismatches, never modifying anything.

Give your AI coding assistant live visibility into the user's browser. Claude sees the actual DOM, console errors, and network timing instead of guessing from screenshots. Drop-in middleware for FastAPI + Flask.

Enables checking whether a Meta, LinkedIn, or Google Ads campaign is structurally able to serve and convert before spending, by inspecting schedule, bid, destination, approval, geo, and budget fields through read-only tools. It also reconciles reported spend across the three platforms against a ledger to surface discrepancies, armed campaigns, and drift.

Transforms single-attempt coding into a multi-attempt, test-validated refinement loop by running your actual test suite and feeding failures back to the LLM as structured directives.

This MCP server verifies that an agent’s claimed tool output matches the actual response returned by a tool to prevent invented or distorted results. It flags mismatches, omissions, and invented fields to ensure downstream logic only processes verified tool data.