Development MCP Servers

Korean company records from official public sources — ~960,000 corporations, official IDs linked, live tax status, English names. REST API + remote MCP server.

Human approval for AI agents: end-to-end encrypted secrets, gated Gmail and GitHub tools, and an approval proxy for any MCP server.

Policy-enforcing gateway for VMware PowerCLI: agents write free-form PowerCLI that is AST-checked and run in a ConstrainedLanguage sandbox, with per-agent rules (vCenters, tag/cluster scopes, time windows, quotas), redaction of sensitive properties, suspicion scoring and SIEM export.

Current WordPress knowledge for AI agents: versions, security, plugin facts, news and docs. Free, read-only, no key.

A/B testing from inside your AI coding agent. Create experiments, read Bayesian results and ship the winner to 100% of traffic from Claude Code, Cursor, Windsurf or Zed.

LiteLLM is the most widely used AI Gateway in the world. Search and read the LiteLLM docs, gateway API reference, and model prices, context windows and capabilities.

Start from a website template with its own database, then edit and publish it from ChatGPT, Claude or any MCP assistant.

Search and read the TurboStarter starter kit docs from any MCP client. Remote server, no auth. Tools: search_docs (ranked hits with URLs and snippets) and read_doc (a docs page as markdown).

What any website is, with evidence: AI website builder traces, AI readiness, popularity, tech stack, DNS records, trust signals and domain history. Eight read-only tools; sign in with OAuth.

Error tracking for AI agents over a remote MCP server (OAuth): read, triage and, when asked, resolve issues; accepts Sentry SDK events and adds cron and uptime monitoring.

Browser automation with evidence. A policy gate checks every action, prompt injections are detected, consequential steps need human approval, and each run compiles into a replayable Playwright test.

Checks an AI coding agent's "I can't" and "done" claims against evidence. Hooks for five agent CLIs plus an MCP server with 5 tools. Python, runs locally.

Test your AI agents and MCP servers for security and safety weaknesses, and re-verify the fixes.

Run App Store Connect from your AI, with a confirm button on everything that costs money. 786 tools in 14 profiles, one-call pricing, offers and submission macros, for Claude, Cursor and Codex.

Verification ledger for agent components: look up the last signed verdict (ALLOW / BLOCK / REVIEW with reasons) for an npm, PyPI, MCP Registry or dsh component before installing. Verdicts are DSSE/Ed25519 in-toto Statement v1, verifiable fully offline and readable by GUAC/cosign. Re-scans supersede rather than overwrite, so a clean-to-compromised flip stays auditable.

Turn links to PDF, Word, PowerPoint, Excel, HTML, CSV or text files into clean Markdown for AI pipelines, one row per document. Ruled PDF tables come out as Markdown tables, and RAG chunks with page numbers are built in. No OCR, so scanned pages come back empty, and password protected files are not supported. It runs as an Apify Actor through Apify's hosted MCP server, so you need an Apify account and API token. Paid per document and per PDF page, 0.004 USD a document and 0.0002 USD a page on the Free plan plus a small start fee. Unofficial tool.

One hosted MCP server for your agent to act in each user's own accounts. Each user signs in to their own tools, and every call runs under that user's limits, waits for approval on risky steps, and leaves a receipt. Works with your own APIs and any remote MCP server.

Remote MCP server (OAuth) to design a data model, generate realistic synthetic data, define metrics with known answers and build dashboards from a description. Free plan within its limits.

Open-core, self-hostable tunnel platform: expose localhost at an HTTPS URL with one command, with TCP/UDP tunnels, request replay and private networking. Its MCP server (zorven mcp) lets Claude/Cursor manage tunnels.

Let Claude Code, Cursor, Codex or any MCP app record product demo videos in Chrome: it drives the tab with a visible cursor while Recordful records, then uploads, shares and reads back the transcript. Also takes styled screenshots.

Remote MCP server for the Truto integration platform. Lets Claude, Cursor and other MCP clients discover and call the Truto admin API (integrations, integrated accounts, unified APIs, sync jobs, logs) from your own workspace. Reads run directly; writes need approval.

Check how any URL's link preview (Open Graph / Twitter card) renders on X, Facebook, LinkedIn, Slack, Discord, iMessage and more, and get a list of concrete issues to fix. Tools: check_link_preview, list_platforms. Free to try without a key (20 calls); paid plans from $9/mo.

Zero-dependency stdio MCP proxy and runtime context firewall: tools/list tree-shaking, 8 MiB crash guard, and secret redaction.

Ask a real, manually approved developer for help right from your AI coding tool (Claude Code, Cursor, Codex). Nothing is sent until you approve. Built for people who made apps with Lovable, Replit, Bolt or Base44 and got stuck.

Remote MCP server for Vibe Codes, a marketplace of apps and code built with AI. Agents can search listings, read one with its honest report, list tags, or start a listing for a seller.

Prompt-injection guard for user prompts that learns from your labels: 50 labels cut false alarms from 15% to 4%. Remote MCP, works keyless.

Plain-English error fixes for Windows, PC games, and code. Search published answers or paste an error. No account or API key.

A free library of ready-made loadouts for AI coding agents, with a read-only MCP server to search and fetch AGENTS.md, CLAUDE.md, rules, hooks, skills and MCP configs.

Official French open data for AI assistants: verify and geocode addresses (BAN), look up French companies (SIRENE, INPI), compute routes and catchment-area statistics (INSEE).

Send a table; get the chart types that fit it, and interactive D3 charts, maps and app widgets as code you own.