A curated collection of AI skills, MCP servers, and workflow tools for content creators.
This is a living collection. Some tools are mine, most are things I've found and vouch for. If something's here, I've used it or someone I trust has.
What's a skill? A markdown file that teaches an AI coding agent how to do a specific task. Drop it into your project, and the agent follows the instructions. No code to install — just a file.
Drop-in instruction files for AI coding agents (Claude Code, Cursor, Windsurf, etc.). Copy the file, point your agent at it, done.
| Skill | What it does | Author |
|---|---|---|
| avoid-ai-writing | Audit & rewrite content to remove AI writing patterns — 61 pattern categories + a 112-entry replacement table. Rewrite / detect / edit-in-place modes, optional voice profiles (casual/professional/technical/warm/blunt), and a zero-dependency detector engine. Installs as a Claude Code / Cowork plugin or a standalone skill. | @conorbronsdon |
| demo-gif-skill | Add a reproducible demo GIF to any repo's README in one prompt. Picks the method (vhs for terminals, Playwright for web apps), writes a committed recording script so the GIF regenerates instead of going stale, optimizes under 8 MB, and embeds it with real alt text. Works in Claude Code and any agentskills.io agent. | @conorbronsdon |
| grokkable-output | Make agent replies parseable in one pass — the first sentence answers the question, detail layers underneath so you can stop reading anywhere. Bans arrow-chain causality, fake structure, and numbers that don't trace to evidence. Ships real unedited before/after outputs. Works in Claude Code and any agentskills.io agent. | @conorbronsdon |
| Skill | What it does | Author |
|---|---|---|
| agent-workspace | Memory and hygiene for agent workspaces — six lifecycle skills (/start, /end, /update, /today, /reconcile, /recover) with a configurable state layout. State files, session logs, daily heartbeats, drift and worktree cleanup. | @conorbronsdon |
| reconcile | Tripwire check for multi-session drift — scans commits and state files for inconsistencies from parallel sessions. Part of agent-workspace. | @conorbronsdon |
| recover | Scan for orphaned worktrees and stale branches after crashes. Read-only by default, cleanup requires approval. Part of agent-workspace. | @conorbronsdon |
| agent-skill-builder | Build agent skills that don't rot. Generates skills from plain-language descriptions with the design decisions most generators skip (invocation control, arguments, context budget), validates them with a bundled checker, and pins its spec knowledge to a CI-watched snapshot. New/review/migrate modes. | @conorbronsdon |
| ssot-check | Single-source-of-truth drift auditor, now tool-backed: deterministic stdlib CLI (45 tests) + skill wrapper + pre-commit hook + GitHub Action. Finds facts hand-copied across docs, builds a manifest of canonical locations, and flags every copy that drifted. | @conorbronsdon |
| repo-audit | Audit a repo against its own code, or get it ready to open-source. For every claim that something is prevented, blocked, or guaranteed, it finds the mechanism and grades it Enforced, Advisory, or Guidance — a warning that prints but does not block is not enforcement. Reads source, tests, hooks and CI before the README; every finding cites file:line and what it could not verify. Six checklist packs, five README templates, optional modules. | @conorbronsdon |
| agent-memory-kit | The curation loop for agent memory — capture, recall, and a read-only curator that finds rot and contradictions before your agent is confidently wrong. Plain markdown and JSON, five slash commands, human-reviewed diffs. | @conorbronsdon |
| agent-skills | Production-tested skills for Claude Code, Codex, Cursor, and OpenCode — session management, multi-session reconciliation, worktree recovery, PR review, and skill scaffolding. The consolidated successor to claude-code-skills. | @conorbronsdon |
| claude-context-os | An operating system for your Claude context — version-controlled files, a self-curating memory, and a /start→/end session loop. Template repo; works in Claude Code and claude.ai. | @conorbronsdon |
| Skill | What it does | Author |
|---|---|---|
| code-review | Multi-agent PR review — orchestrates Copilot + parallel subagents (adversarial, operational, reference-comparison) sized to PR risk. Catches architectural P0s that single-pass review misses. | @conorbronsdon |
| eval-integrity | Audit an LLM benchmark repo for credibility practices — 36 scoreable sub-checks across seven dimensions, a JSON result schema for CI, and three audit fixtures. Answers: would your published numbers survive an adversarial reviewer? | @conorbronsdon |
| readme-audit | Audit a README against what the software actually does. Reads the CLI help, schemas, and parity tests first, then reports P1/P2/P3 findings with the evidence checked and the sources it couldn't verify. Fails a README that claims enforcement without naming the mechanism doing it. Read-only unless you ask for a rewrite. Part of nnennandukwe/skills. | @nnennandukwe |
| software-build-plan | Turn one ticket into an implementation-ready plan — contracts, file seams, commit sequence, test plan — before any code is written. Its pre-PR gate requires a second review from a different model family, pinned by model id, because a second pass from the same family shares its blind spots. Part of nnennandukwe/skills. | @nnennandukwe |
| failure-path-testing | Write the failure test first — start from the transition that must stay blocked, the gate that must refuse, the empty config that must fail loudly instead of resolving to a wrong default. Part of nnennandukwe/skills. | @nnennandukwe |
| Skill | What it does | Author |
|---|---|---|
| angel-diligence | Pre-investment research and deal-memo generation. Parallel web research with strict citation rules; separates verified facts from claims; ends in a verdict scaffold, never an invest/pass call. | @conorbronsdon |
| guest-circuit | Map a prospective podcast guest's appearance circuit — where they've been, their stump speech, and the unclaimed angle for your show. Uses Podcast Index when configured, web search otherwise. | @conorbronsdon |
What's MCP? Model Context Protocol — a standard way to connect AI agents to external tools and data (calendars, databases, APIs). Think of it as a USB-C port for AI.
MCP servers give AI agents access to real tools and data.
| Server | What it does | Author |
|---|---|---|
| Transistor-MCP | Full Transistor.fm API access — episodes, analytics, transcripts, show management. | @conorbronsdon (fork of @gxjansen) |
| substack-mcp | Read posts, manage drafts on Substack. No publish or delete by design — safe for agent workflows. | @conorbronsdon |
| podcastindex-mcp | Podcast Index API — search by person/topic, trending podcasts, feed health checks, cross-platform episode discovery. | @conorbronsdon |
| apple-podcasts-mcp | Owner-side Apple Podcasts analytics over the Podcasts Connect Reporter protocol — plays, unique and engaged listeners, followers, per-episode listening. Not the public iTunes catalog. Read-only. | @conorbronsdon |
| op3-mcp | OP3 (Open Podcast Prefix Project) analytics — downloads over time, listener geography, app share, per-episode breakdowns. Read-only by design. | @conorbronsdon |
| postlint-mcp | Check a draft against a platform's real character limits before you post — X bills every URL at 23 via t.co and weights non-Latin 2x, Bluesky counts 300 graphemes, Mastodon counts grapheme clusters. Every limit carries a primary source. | @conorbronsdon |
| gsc-mcp | Google Search Console — search analytics, sitemap submission, URL inspection. Read tools work on a read-only scope; writes ask for the full one. | @conorbronsdon |
| yt-analytics-mcp | Owner-side YouTube Analytics — watch time, traffic sources, audience retention, geography, and playlist metrics, plus an episode-race tool that re-indexes videos to days-since-their-own-publish so a new episode compares against the last few at the same age. The owner-only half the Data API can't reach: no API key gets you retention or watch time. Nine tools, all reads, channel==MINE hardcoded. |
@conorbronsdon |
| Server | What it does | Author |
|---|---|---|
| gws-mcp-server | Google Workspace access — 39 curated tools across Gmail, Calendar, Drive, Sheets, Docs, and Tasks. Built on the official gws CLI. |
@conorbronsdon |
Publishing and scheduling across social platforms. Read the gating notes below before you plan a workflow around any of these — the API is rarely the hard part.
| Server | What it does | Author |
|---|---|---|
| X MCP | Official hosted MCP at api.x.com/mcp. Full-archive search, post and engagement lookup, user timelines, bookmarks, trends, and X Articles drafting. OAuth 2.0 via the xurl bridge for user-context writes, or an app-only bearer token for read-only. |
@XDevelopers |
| Postiz | Self-hosted scheduler covering 32 networks including TikTok, Instagram, Threads, Bluesky, and Mastodon. A documented public API (POST /posts, POST /upload, GET /integrations) plus a Node SDK make it scriptable. You still supply your own developer credentials per platform — the compose file takes TIKTOK_CLIENT_ID, THREADS_APP_ID, LINKEDIN_CLIENT_ID and the rest, so it removes the integration work but not the app-review work. Runs on 2GB RAM / 2 vCPU. No MCP server. AGPL-3.0. |
@gitroomhq |
| atproto-mcp | Bluesky and the wider AT Protocol — 51 tools covering posts, threads, feeds, search, lists, DMs, and moderation. Runs read-only with zero config (npx atproto-mcp), or add an app password for writes. Every tool declares MCP annotations and destructive operations are flagged, which is rarer than it should be. Built on the official @atproto/api and MCP SDK, with rate limiting. MIT. |
@cameronrye |
| meta-mcp | Instagram Graph API and Threads API in one server — 57 tools on Graph API v25.0. Publishes photos, videos, reels, stories, and carousels to Instagram; text, polls, GIFs, and carousels to Threads. Includes insights, comment management, and a cross-posting prompt. | @mikusnuz |
| Typefully | Hosted drafting and scheduling for X, LinkedIn, Threads, Bluesky, Mastodon, and Substack Notes, with an official REST API, hosted MCP, and Agent Skill. Its OAuth connections live with the service, so the per-platform app-review gating below stops being your problem: an agent drafts through the API and your connected accounts publish on schedule. Planned drafts sit on the calendar inert until confirmed, which makes a clean review gate for agent-written copy, and media upload works over the API (tested with 15-25MB video clips). Publish volume is tier-gated. | Typefully |
What actually gates you. Posting is the last mile and the easy part. The work is upstream, and no scheduler removes it — self-hosted tools take your credentials, they don't lend you theirs:
- TikTok — personal accounts cannot use the Content Posting API. You need a business or developer entity, plus app review with a privacy policy and a demo video of your OAuth and upload flow. Budget 5–10 business days for approval.
- Instagram — Graph API publishing needs a Business or Creator account and a linked Facebook Page. Verify which account type your workflow requires before you commit; the answer has moved more than once.
- Threads — free API, and since September 2025 it no longer requires a linked Instagram account. Production access still waits on Meta App Review.
- LinkedIn — the most restrictive of the major platforms. Publishing runs through the Community Management API, which needs a registered company, a verified Page, and a two-tier app review. There is no quick path for individual developers.
- Bluesky — the exception. App passwords, no review, no paid tier, open protocol. If you want one platform automated this week, it's this one.
Known gap, narrowed. Postiz ships no first-party MCP server — as of 2026-07-30 its repo carried one internal start.mcp.ts and docs.postiz.com/public-api documented no MCP surface. Third-party wrappers do exist, which the earlier flat "no MCP server for Postiz" here got wrong. The most developed is postiz-mcp — on npm, MIT, 22 tools, writes gated behind POSTIZ_ENABLE_WRITE and deletes behind a second flag plus confirm: true. It stays out of the table above because nobody here has run it against a live instance, and the bar for this list is having used a thing. Treat it as a lead, not a recommendation.
On finding the Bluesky one. Searching GitHub for a Bluesky MCP returns a dozen-plus results, and sorting by stars is actively misleading here — the most-starred was last touched in April 2025. atproto-mcp sits at single-digit stars and is the only one that's actually maintained, published to npm, and annotated properly. Worth remembering when a category looks abandoned: it may just be badly sorted.
If you're building with agents, you eventually need to measure them. Tools for that:
| Project | What it does | Author |
|---|---|---|
| cot-bench | Open agent evaluation leaderboard. Three judges score every scenario (two open-weight, one frontier reference) across CLEAR-aligned metrics: efficacy, cost, reliability, latency. Rubrics are code, and every raw judge score is published so you can audit the results. | @conorbronsdon |
| podcast-benchmark | Benchmark any podcast against its peers using only public data — catalog depth, cadence, transcript availability, feed hygiene. No download estimates, every number sourced and timestamped. | @conorbronsdon |
Standalone tools that run in the browser. No agent, no install.
| App | What it does | Author |
|---|---|---|
| track-finder | Paste a tracklist or a YouTube playlist URL and get search links for every track on SoundCloud, Spotify, YouTube, Apple Music, or Beatport. Progress saves in your browser. Built for DJs and playlist builders. | @conorbronsdon |
Most skills follow the agentskills.io standard and work across tools. For Claude Code specifically:
- Download the skill file (usually
SKILL.mdor similar) - Place it in your project directory or reference it in your CLAUDE.md
- Invoke it — most skills include slash command instructions
Skills in the agentskills.io format work with Cursor, Windsurf, Cline, OpenHands, and 40+ other tools. Check each skill's README for tool-specific setup.
Each MCP server has its own install instructions. Generally:
- Clone the repo
npm install && npm run build- Add the server to your Claude Code or Claude Desktop config
See Anthropic's MCP documentation for setup details.
This is a curated list, not a dump. If you've built or found a tool that's genuinely useful for content creators, knowledge workers, or anyone who ships ideas for a living:
- Open an issue describing the tool and why it's worth including
- Or submit a PR adding it to the right section
Quality bar: you've actually used it, not just bookmarked it.
Maintained by Conor Bronsdon. I host the Chain of Thought podcast covering AI infrastructure, developer tools, and how practitioners actually use this stuff. Many of these tools were built or discovered through conversations with guests on the show.
New to AI tools? Check out AI Learning Resources — a curated path from "what is AI?" to building your own workflows.
If you want to go deeper on any of these topics, the podcast is a good place to start.
This is an independent personal project, not affiliated with, sponsored by, or endorsed by any company. All views expressed are my own.
MIT (see LICENSE).