Guide library · 80 real routes

Implementation guides that lead to the next right step.

For MCP operators, agent builders, and runtime troubleshooters—choose a setup, fix, comparison, or workflow guide without scanning an unrelated directory.

SETUP 54 · TROUBLESHOOTING 2 · COMPARISON 5 · WORKFLOW 19

Recommended first

Check the MCP security boundary before connecting a new server.

Verify source identity, permission scope, exact-client behavior, credentials, logs, and rollback before enabling writes.

Open security checklist

This checklist verifies

  1. 01Source identity and the exact package route
  2. 02Permission scope, credentials, and data exposure
  3. 03Read-only proof, logs, and a bounded rollback path

Full guide library

Browse all 80 guides without opening 80 tabs.

Every real route is visible below. Filter by outcome or search the title, stage, summary, and best-use guidance.

Showing 80 of 80 guides

54 guides

Setup

Install, configure, and validate a new agent or MCP environment.

WPVibe MCP Setup: Connect WordPress and Check PermissionsSeparate client sign-in from WordPress authorization. Check the target site, read/write permissions and rolling tool-call allowance before editing.Site and permission checksSuno MCP Setup: Clients, Beta Access and Credit ChecksCheck beta admission, connect the intended Suno account and verify credits and library results before retrying a music request.Account and song checksAnthropic sandbox-runtime: Test Agent Filesystem and Network LimitsUnderstand sandbox-runtime, prepare a local macOS or Linux fixture and check allowed writes, denied reads and network access before wrapping agent commands.First verified workflowGoogle Workspace CLI (gws): Setup and a Read-Only Drive TaskInstall gws, choose a narrow OAuth setup, inspect the Drive schema and run a small file-list query before giving an agent broader Google Workspace access.First verified workflowMicrosoft Skill Recorder: Turn a Demo into a Reviewed SKILL.mdPrepare Skill Recorder, record a small synthetic workflow, review the generated SKILL.md and test reuse while understanding cloud analysis and Copilot requirements.First verified workflow1Password for Agents: Set Up Claude Login and Check PermissionsSet up the documented 1Password and Claude integration on Mac, understand plan requirements and verify a first login without pasting secrets into chat.First verified workflowMCP Tasks and Elicitation: Long-Running Tools in Claude CodeCheck Claude Code MCP Tasks support, follow a long-running job, answer elicitation requests, cancel deliberately and verify the final result.Long-running MCP toolsBuild an MCP Server for Your App with SupabaseExpose public or user-specific app tools through Edge Functions. Configure OAuth and RLS, verify returned data and understand usage costs.Build your app MCPMirror CLI Setup: Model Providers, Worktrees and RecoverySet up Reflection Mirror CLI, select a model provider, verify a first repository task and run parallel worktrees with clear approval and recovery boundaries.Provider and worktree setupDeepSeek Search & Fetch: Claude Code and Codex SetupConfigure native search, inspect page-content responses, verify citations and separate tool fees from model tokens.Search and source verificationClaude Platform OpenTelemetry: API and SDK tracingChoose API, SDK or combined tracing, keep Metadata only and avoid duplicate token counts.API observabilityClaude Code Desktop: Parallel Sessions and WorktreesCreate two Claude Code Desktop sessions, verify worktree isolation, review diffs and preserve unfinished work before archiving.Desktop workflowPerplexity SPACE Sandbox: Setup, Isolation and Safe RecoveryChoose SPACE or a managed sandbox, check Linux/KVM prerequisites, verify command output and recover without duplicating sandbox work.Agent code executionClaude Code Review Setup: GitHub, Review Levels and CostEnable Claude Code Review for a GitHub repository. Compare Quick, Standard and Deep, permissions, PR triggers, billing and acceptance checks.Repository reviewConnect a Custom MCP Server to ChatGPT: Setup & PermissionsAdd an existing MCP server to ChatGPT as a plugin. Follow the current URL or Secure MCP Tunnel flow, OAuth, tool refresh and read/write approval boundaries.Connect your MCPAntigravity Models by Plan: Gemini 4 Argon & Nov 2 MigrationCompare Antigravity models by plan: Gemini 4 Argon access, supported clients, Claude 5.5 eligibility and the November 2 Claude 4.6 / GPT-OSS migration. Source-checked 14 Oct 2026.Model access and migrationPi Durable Agent Harness: MCP, Crash Recovery & Cloudflare PiHarnessChoose normal Pi or experimental Pi Durable. Design MCP exposure, persistent state and safe retries, with Cloudflare PiHarness as an optional beta runtime.Agent recoveryHedra MCP & CLI: Media Jobs from Claude, ChatGPT or CodexConnect Hedra through MCP or CLI, check credentials and model inputs, submit a media job, and verify the output. Understand wallet, storage and retry boundaries.Agent media jobsChatGPT Sites Plugins: Host MCP Tools on a SiteAdd MCP tools to a ChatGPT Site, publish its associated plugin, install and connect it, and share workspace access without confusing Site and app permissions.Site-backed toolsGitHub Copilot Computer Use: Setup & App PermissionsEnable GitHub Copilot computer use on macOS or Windows in the CLI or app. Understand /computer commands, saved app approvals, organization policies and safety limits.Desktop workflowsChatGPT Plugin Extensions: Sidebar, Panels, Files & SettingsAdd optional UI to an MCP-backed ChatGPT plugin: sidebar apps, conversation panels, settings and file handlers, with SDK guidance and current rollout limits.Plugin developmentMCP Events for ChatGPT Plugins: Subscriptions & WebhooksImplement ChatGPT MCP Events with protocol 2026-07-28, event discovery, subscriptions and signed webhooks. Covers filters, retries, revocation and loop prevention.Plugin developmentHiggsfield Skills and MCP: choose your setupChoose between creative-app workflows, repository skills and cloud MCP; check local tools, credits and access before starting.Creative workflowsMeta Muse Guide: Access, Connectors and a First TaskUnderstand Meta Muse availability, connector permissions, purchase approvals and a limited first task. Announced features are separated from verified use.First workflowAgent-Reach Setup: Check Channels and Read a First SourceInstall the official Agent-Reach CLI, understand its read-only default checks, and verify one public-source task without assuming every channel works.First workflowCubeSandbox Guide: Host Prerequisites and a First SandboxUnderstand CubeSandbox as agent execution infrastructure, choose the documented deployment path and verify a template, disposable task and cleanup.First workflowWeKnora Setup: Import a Document and Verify a Cited AnswerStart a small WeKnora knowledge-base trial with Docker Compose, check model configuration and document parsing, and verify answer citations against your source.First workflowZCode Setup: Choose Desktop, Web or TerminalUnderstand Z.ai ZCode, choose a local interface, check source-build prerequisites and run a narrowly scoped first repository task.First workflowDeepSeek Harness Setup: Model, Workspace and First TaskStart with the official dsh Web entry, configure a model, select a workspace and verify a small task. Developer-preview limits explained.First workflow3dviz-pro-max Setup: Load the Skill and Verify a First SceneInstall the complete 3dviz-pro-max skill folder, confirm agent discovery, and evaluate a small 3D scene without confusing recipes with tested outputs.First workflowTencent BrowserSkill: setup and first workflowConnect an agent to Chrome or Edge, understand browser access, and start with a copyable read-only task.First connectionAgent Skill MCPsUtility for tracking Claude Code token usage, visualizations.publishedMaster Wiki Creation with pi-llm-wikiDiscover how pi-llm-wiki turns data into interlinked knowledge bases through Obsidian, MCP, and OKF.ReleasedMCP Inspector testingTest and debug any MCP server with the official Inspector web UI, CLI, and TUI before adding it to a client config.Server testingCodex CLI skills directorySource-linked Codex skills for PR feedback, CI debugging and browser work, with installation paths and a SKILL.md template.Codex setupHermes Agent quickstartInstall and configure Hermes Agent with CLI setup, chat integrations, auto-skill generation, and VPS deployment options.Agent bootstrapHow to create an AI agent skillCreate and reuse a SKILL.md across compatible agents: shared format, client paths, environment requirements and a copyable text-only example.Skill authoringCursor skills hubChoose Cursor-ready MCP servers, rules, and workflow skills without overloading the editor or permission surface.Editor setupMCP agent skillsUnderstand how MCP servers become repeatable agent workflows with validation, permission checks, and rollout rules.Agent toolingMCP install examplesUse source-aware command examples for filesystem, memory, Git, and browser MCP servers with safer defaults.Install proofMCP server security checklistVerify source identity, permission scope, exact-client behavior, read-only proof, credentials, logs, and rollback before enabling writes.Security gateBest MCP serversCompare high-utility MCP servers by category, install surface, assistant fit, and operational risk.Discovery@modelcontextprotocol/server-everything npxVerify the official reference-server package, npx command shape, Claude Desktop config, VS Code config, and Windows wrapper before testing.Command validation@modelcontextprotocol/server-filesystem npxInstall the official filesystem MCP server with npx, allowed directories, Claude Desktop config, VS Code config, Windows wrapper, and safety checks.Filesystem setup@modelcontextprotocol/server-memory npxInstall the official Memory MCP server with npx, Claude Desktop config, VS Code config, Windows wrapper, MEMORY_FILE_PATH, and local knowledge graph safety checks.Memory setupmcp-server-git uvxCorrect stale @modelcontextprotocol/server-git npx snippets and configure the official Git MCP server with uvx, repo paths, Claude Desktop, VS Code, Docker, and write-risk checks.Git setupmcp-server-fetch uvxCorrect stale @modelcontextprotocol/server-fetch npx snippets and install the official Fetch MCP server with uvx, Claude Desktop, VS Code, Docker, pip, and SSRF checks.Network fetchchrome-devtools-mcp npxInstall Chrome DevTools MCP with npx, Codex Windows config, first performance trace, browser-state safety checks, and Playwright comparison.Browser debuggingMCP desktop extensionsUnderstand DXT legacy naming, current MCPB bundles, Claude Desktop extension install paths, and safe rollout checks.PackagingMCP Apps directoryBrowse MCP clients, apps, reference servers, and official-source links before trusting a package name.VerificationNeonCtl MCP init guideInstall and bootstrap the NeonCtl MCP workflow with predictable local setup steps.Bootstrapnpx neonctl@latest initUse the current Neon CLI init flow with MCP clients, agent skills, and safety boundaries verified for 2026.BootstrapMoltBook install guideStep-by-step setup notes for MoltBook environments and MCP usage.EnvironmentEvoMap MCP server setupConfigure EvoMap as a real MCP service with validation checkpoints.Integration

2 guides

Troubleshooting

Recover a broken or drifting environment without resetting blindly.

Claude Code idle compaction: disable, configure and verifyKeep ordinary auto-compact while controlling idle summarization. Includes token thresholds, local settings, hooks and recovery notes.Context configurationMoltBook skills not found fixTroubleshoot missing skill discovery and indexing issues inside MoltBook.Recovery

5 guides

Comparison

Choose a stack, framework, coding surface, or operating model.

Claude Code vs Codex vs CursorCompare three AI coding assistants across MCP support, pricing, autonomous coding, context handling, and workflow fit.Coding assistant selectionHermes Agent vs OpenAI CodexCompare self-hosting, provider choice, coding surfaces, permissions, billing boundaries, and a safe first trial.Workflow selectionHermes Agent vs OpenClawCompare two open-source agent frameworks across architecture, skills, memory, security, pricing, and deployment fit.Framework selectionBest AI skill stacksCompare eight curated MCP server combinations for development, DevOps, content, data, security, and SaaS workflows.Stack selectionEvoMap vs OpenClawCompare positioning, workflow fit, and deployment tradeoffs between two agent stacks.Selection

19 guides

Workflow

Turn useful agent patterns into repeatable team operations.

Harmar MCP: connect, caption and exportChoose remote MCP or API credit, prepare a supported video and check subtitle or burned-in output, queue recovery and retention.First caption workflowOriginal Agent Workflows: Static Release PreflightDownload our original Skill and local checker, try a practice fixture and inspect the real six-guide release replay.Inspect and reuseOpenAI Decisions API: Predicate, Choice & ScoreChoose Decisions or Responses, understand typed outputs and confidence, and design routing with human review.Typed routingGoogle Antigravity /plan: review before codingExplore the codebase, review an Implementation Plan and choose when to proceed. Includes CLI plan mode and current review-policy differences.Plan and reviewCodex /goal: run a persistent coding objectiveSet a measurable Goal, manage its lifecycle and use evidence to decide when iterative coding work is complete.Persistent objectivesSubmit a Claude plugin to the directoryPrepare a plugin bundle, use the official submission portal, follow review and distinguish real usage from sample analytics.Plugin publishingMastra Factory: issue to reviewed PRA practical walkthrough with human checkpoints, setup prerequisites and a copyable issue template.Team deliveryAgent Skills Hub guideDefinition, safety checklist, and navigation path for developers comparing agent skills, MCP servers, and workflow guides.OrientationHermes Agent skills lifecycleCreate, review, install, test, approve, and share Hermes Agent SKILL.md files with source and permission checks.Skill operationsMCP prompt examplesCopy safer prompt patterns for repository review, filesystem docs, browser QA, and database triage with approval checkpoints.Prompt designAgent workflow case studiesPreview QA, repository preflight, and database investigation workflows with prompts, evidence, and human approval points.Case studiesSkills library for agentsStructure an internal agent skills library with categories, metadata, safety limits, versioning, and review cadence.Library designAI agent memory systemsCompare file-based memory, SQLite, vector databases, RAG, and hybrid architectures with implementation examples.Memory architectureBest AI prompt librariesCompare official, community, framework, and marketplace prompt collections for Claude, GPT, and multi-model workflows.Prompt discoveryWeekly trending AI skillsExplore AI agent skills and MCP projects with repository signals, source checks and practical setup guidance.Signal reviewMCP browser automationChoose between local Puppeteer/Playwright servers, managed browser sessions, and search-first workflows.BrowserEvoMap workflow patternsProduction workflow patterns for repeatable orchestration and execution loops.Scale-upEvoMap prompt memory syncKeep prompts, memory, and runtime coordination aligned across agent sessions.OperationsEvoMap OpenClaw bounty monetizationA practical guide to monetization angles around OpenClaw-style ecosystems.Go-to-market

One job, one next guide

Open the guide that matches the decision in front of you.

All eight existing decision routes are retained below, so the richer page does not hide useful operator guidance.

I need a fresh local setup that teammates can reproduce.npx neonctl@latest initIt is the clearest route for validating first-run artifacts, prerequisites, and post-init checks.I need to know which MCP server or client source is trustworthy before install.MCP Apps directoryIt separates official and community sources so discovery does not turn into package-name guessing.I need browser control for QA, scraping, screenshots, or authenticated portal checks.MCP browser automationIt maps local Playwright/Puppeteer and managed browser choices to the right risk level.I need to package a local MCP server for one-click desktop install.MCP desktop extensionsIt explains DXT legacy naming, current MCPB bundles, install friction, admin controls, and rollout safety checks.I already know I need Neon MCP, but I want the broader rollout logic.NeonCtl MCP init guideIt frames setup as a team baseline instead of a one-machine command.The install finished, but skills are not showing up correctly.MoltBook skills not found fixIt targets discovery and indexing breakpoints instead of generic install steps.I need to write the first SKILL.md, not browse another catalog.How to create a skillIt owns the authoring contract: trigger, install path, permission check, and first reversible proof.I need to compare two agent stacks before committing.EvoMap vs OpenClawIt covers positioning, workflow fit, and deployment tradeoffs in one place.The tools are installed, but team execution is still inconsistent.EvoMap workflow patternsIt focuses on repeatability and operating discipline rather than onboarding.

Common implementation questions

Answers that help you choose the right next guide.

What is the difference between the guides hub and the main skills directory?

The skills directory helps you discover tools and capabilities quickly. The guides hub is for implementation detail, rollout decisions, troubleshooting, and operating patterns after discovery.

Which guide should I read first for Neon MCP onboarding?

Start with the npx neonctl@latest init guide if you need the exact CLI setup path. Move to the NeonCtl MCP init guide when you need the broader rollout policy and team baseline.

Are these pages only for one editor or one agent runtime?

No. The hub is organized by outcome: setup, troubleshooting, comparison, and workflow. Most guides call out where editor-specific or runtime-specific behavior matters.

How should teams use this hub during implementation?

Use the start-here paths first, open one guide that matches the current bottleneck, and only add a second guide when the problem changes from setup to troubleshooting or from implementation to operating model.

When should I leave the guides hub and go back to skills pages?

Return to the skills directory when you are exploring tooling options. Stay in the guides hub when you already know the tool and need steps, guardrails, validation, or workflow decisions.

Where should I start for MCP server discovery or browser automation?

Start with the MCP Apps directory when source verification matters, the Best MCP Servers guide when you need a shortlist, and the MCP browser automation scenario when the work involves page rendering, screenshots, or web task execution.

Need a safer first step?

Open the MCP security checklist before connecting a new server.

Review permissions, credentials, transport, data exposure, and rollback before installation.

Open security checklist