Home / AI agent tools

AI Coding Agent Tools & Troubleshooting Hub

Use working browser tools, exact-error guides and primary-source update tracking for Codex, Claude Code, GitHub Copilot, Antigravity, MCP, browser agents and AI-generated applications.

Last verified: August 2, 2026. Product behavior changes quickly; every time-sensitive section links to the product owner.

Paste an exact error

Route a short log locally to one of 21 reviewed Claude, Codex, MCP, ComfyUI, Gemini or Vite resolution paths.

Run local error triage → My MCP config fails

Validate JSON or Codex TOML, catch secret leaks, and convert it for the target client.

Run MCP Config Doctor → I need current product behavior

Start with release notes and official documentation before social summaries.

View official watch →

A repeatable coding-agent diagnostic workflow

  1. Capture exact evidence. Preserve the full error, command, agent version, operating system, working directory and timestamp. Remove tokens and private data before sharing.
  2. Classify the boundary. Decide whether the failure is local process startup, filesystem permission, authentication, MCP transport, provider quota, browser session or generated application behavior.
  3. Check the product owner. Verify current installation and platform support in official documentation. A six-month-old command may now be wrong.
  4. Reduce to one failing action. Disable unrelated MCP servers, plugins and custom instructions; reproduce using one repository and the smallest required permission.
  5. Inspect permission and identity separately. A successful login does not prove that a process token, OAuth scope, repository policy or filesystem ACL authorizes the attempted action.
  6. Verify the fix. Repeat the original action, record the changed variable and add a rollback. Do not call a workaround a root-cause fix without evidence.

Browser-side tools for agent workflows

MCP Config Doctor & Converter

Diagnose configuration before chasing runtime failures.

  • Claude, Cursor, VS Code, Codex and Gemini formats
  • JSON ↔ TOML conversion and exact wrapper checks
  • Hardcoded secret and transport warnings

Official MCP Registry Browser

Search live Registry metadata, then review identity, artifacts, credentials, permissions and maintenance before connecting.

  • No fake star rankings or static server count
  • Packages, remote transports and declared secret fields
  • Copyable registry ID and server.json metadata

llms.txt Workbench

Build proposal-shaped Markdown, validate an existing file, or turn Sitemap XML into candidates for manual curation.

  • Local parsing with no URL fetch or content upload
  • Line-level structure, duplicate and path warnings
  • Explicit Google boundary: not a ranking signal

HTML Preview & Safety Inspector

Preview AI-generated HTML with scripts and network blocked by default.

  • Document structure and accessibility signals
  • Scripts, event handlers and unsafe URLs
  • Origin-isolated iframe and restrictive CSP

MCP Token Budget Analyzer

Estimate schema context cost before exposing a large tool catalog.

  • Find verbose descriptions and duplicated shapes
  • Compare tool count with usable context
  • Keep schema data in the browser

MCP OAuth Debugger

Walk through redirect, state, callback and token assumptions without submitting secrets.

  • Separate client, authorization server and resource server
  • Identify callback and state mismatches
  • Generate a redacted checklist

Apps SDK Manifest Validator

Check common manifest and metadata problems before testing an app integration.

  • Required fields and URL shape
  • Widget and tool metadata checks
  • Local validation only

Browser Agent Debug Checklist

Distinguish CDP, session, accessibility-tree and page-state failures.

  • Connection and target checks
  • Visible versus semantic page state
  • Evidence for reproducible bug reports

AI-Generated App Production Checklist

Review auth, data, secrets, dependencies, deployment and rollback before launch.

  • Production gate rather than style score
  • Explicit unknown and not-tested states
  • Downloadable review output

High-intent troubleshooting paths

Codex on Windows

Process identity and ACL errors need different fixes.

Claude Code

Separate provider capacity, authentication and local configuration.

MCP and OAuth

Verify the server record and configuration first, then trace process environment, throttling and token lifecycle independently.

ComfyUI and Vite builds

Route high-demand GPU and import failures using the first causal line and a measurable verification gate.

Official product-update watch

OpenAI · June 2, 2026

Codex plugins, annotations and Sites

OpenAI's current direction expands Codex beyond code into role-specific plugins, in-place annotations and shareable sites. Build help pages around concrete workflows and permission problems, not speculative feature lists.

Official announcement →

GitHub · July 8, 2026

Agent browser, sessions and MCP OAuth

GitHub documents agentic browser tools, session sync, larger context support and preregistered MCP OAuth credentials in current VS Code releases.

Official changelog →

Google · May 19, 2026

Gemini CLI to Antigravity CLI migration

Google announced the consumer transition to Antigravity CLI and a June 18 service cutoff for affected Gemini CLI and Code Assist access. Troubleshooting should confirm which CLI is actually supported first.

Official announcement →

Cloudflare · July 22, 2026

MCP exposure controls and Code Mode runtime

Cloudflare's Agents SDK update reduces repeated schema conversion, adds MCP exposure controls and exposes direct Code Mode runtime methods. This creates durable demand for schema-budget and approval-boundary diagnostics.

Official changelog →

Individually reviewed workflow guides

Only guides with a concrete workflow, current source check and prior search demand are promoted here. The bulk blog archive remains excluded from search during quality recovery.

Ruflo install: plugin, CLI or MCP?

Choose the smaller evaluation path, use current commands and inspect what initialization changes.

Read verified guide → Claude Code HTML output workflow

Prompt for a reviewable artifact, preview with scripts blocked, then inspect interaction separately.

Read safety workflow → Ollama + Open WebUI Docker setup

A current local AI stack with CPU/NVIDIA paths, verification, backup and connection diagnostics.

Open setup guide →
EasyTool publishing rule: a new agent page must contain a working tool, a reproducible error, a tested workflow or primary-source analysis. A product announcement by itself is not enough.

Security boundaries for agent-operated tools

  • Give agents the minimum repository, filesystem, network and account access needed for the current task.
  • Keep credentials out of prompts, logs, screenshots, URLs and downloaded diagnostic reports.
  • Treat generated commands as proposals until scope and targets are reviewed.
  • Require explicit review for destructive actions, permission changes, deployments and messages to third parties.
  • Verify generated applications for authentication, authorization, input validation, secret handling, dependency risk, monitoring and rollback.

Have a concrete failure? Paste the exact error into the local reviewed router, or browse the indexable tools directory.