Development MCP Servers

Self-hosted daemon that lets coding agents (Claude Code, Codex, OpenCode, OpenClaw, Pi, and A2A-compatible agents) communicate and delegate tasks to each other, on one machine or across many. Existing coding agents connect through MCP, while daemons communicate using the open A2A protocol. Each developer controls which peers their agents can communicate with.

Validate and normalize postal codes for 252 countries and territories over MCP and REST. Tools: validate_postcode and list_countries. Format check only — it does not verify that an address exists or is deliverable. Free tier 2,000 calls/month; Basic $9/month for 50,000; Pro $29/month for 500,000.

Let Claude Code, Codex and any AI agent use the Chrome you're already signed in to. Open-source CLI + MCP server + Chrome extension (native messaging, no debug port); stealth by default. Docs: https://chrome-use.leeguoo.com/en/

Query analyzed WebRTC sessions from rtcStats: list calls, get a full analysis, check quota.

Delegate scoped coding tasks from Codex to native DeepSeek Harness through a local stdio MCP server with explicit workspace isolation.

Finds injection bugs by following attacker data across files, checks each path with a separation logic solver, and verifies fixes. Run with uvx frame-mcp.

Test and monitor voice and chat AI agents from your AI assistant: run simulated calls, score live calls and manage evals.

Find open-weight AI models that fit your hardware, with weight sizes, memory and sources.

Search the BANKSapi Open Banking documentation and fetch exact OpenAPI operations while you build your integration. Read-only, public docs, no account needed.

An open source MCP server that gives your model randomness from outside it, so one prompt gives different answers that still do what you asked.

It helps users to use VS Code's Continue extension non existing features like analyze image files. fetch web images, lots of automatisations that helps hte agent code more effectively.

I built forge-delegate, an MCP server that lets Claude Code, Codex or opencode hand coding tasks to another model. Think writing tests, renames across files, fixing lint errors, boilerplate, or reading through a big codebase to answer one question. The other model runs as a full agent through opencode. It reads the files and makes the edits itself. Your main agent only gets back a short summary, so its context stays small and it can focus on the decisions. Second opinions with ask_model. Your agent can also ask a different model a direct question, with no agent loop and no file access, and get a fast answer back. It's useful when you want another model to check a design decision, a bug theory or a tricky piece of code before your agent commits to it. Every model has its own blind spots, and a different one often catches what yours missed. Your agent writes the question with whatever context is needed, then weighs the answer itself. Just tell it something like "ask DeepSeek what it thinks" or "get a second opinion from Kimi", naming any provider you've logged into opencode. Any provider opencode supports. forge-delegate doesn't hold its own API keys. It uses whatever you've set up in opencode. So DeepSeek, Kimi, OpenAI, Gemini, OpenRouter, custom OpenAI-compatible endpoints, and local models through Ollama all work. Log in once with opencode auth login <provider> and that provider's models are available to delegate to. opencode also ships free models that need no account. You can set a default model, save presets for different kinds of tasks, change either at any time, and run several delegations in parallel in the background. Setup: npx -y forge-delegate setup It walks you through picking a default model and which agents to register with. Things to know: - By default, the MCP instructions and skill steer your agent toward cheap or free models. It sticks to your default model and won't switch to a pricier one on its own. If you want a stronger model, set it as the default or ask for it. - Delegated agents can edit files and run commands without asking by default. You can turn that off, but only point it at repos you trust. - Smaller models make more mistakes, so review what comes back. - I mostly use it with Claude Code on a Mac. Codex and Windows should work, but they've had less real-world use, so issues are welcome. MIT licensed: https://github.com/MuhamadBarzani/forge-delegate Feedback welcome, especially on which tasks it handles well and where it falls short.

7Maps checks every remote MCP server in the official registry and signs every check. Agents ask which server works and which tool fits, before connecting.

Compare first-text latency, total latency, and estimated output throughput across provider APIs from a local stdio MCP server. Uses your own API keys.

Compiles agent skills into deterministic pipelines and exposes each compiled pipeline as an MCP tool

Discover and run XML conversion workflows for JSON, CSV, XLSX, HTML, TXT, PDF and YAML from MCP-compatible AI clients.

Convert HTML to PDF, DOCX, JPG, PNG and XML, or create HTML from XML, DOCX, XLSX and XLS.

Remote MCP server for adding Giggy voice AI to compatible clients: free Batch TTS/STT and aggressively priced streaming TTS at $10 per million characters and STT at $0.08 per input audio hour. API key required.

Hourly security advisories for JS, AI and web3 packages: ranges, fixes and exploit conditions.

Verifiable action receipts + reproducible GDPR checks for any agent. Local-first, no telemetry.

Skill librarian: agents describe a task and get the few matching skills from a large local skill collection, then report whether they worked.

Deterministic AST security scanner for Python and JS/TS (65 rules, no LLM, ~0.2ms) with a PASS/WARN/BLOCK execution gate and auto-remediation diff, pre-flight simulation for Base and Solana transactions, and an anti-bot scraper oracle. Paid per call via x402 (USDC on Base or Solana). Remote MCP endpoint, no install.

Search and read local Ruby and gem documentation matched to your project versions. Local stdio server; POSH agent-tools license required.

Ask your AI agent "is AWS down?" Real-time status, incidents and an overall health score for 41 backbone services. Hosted, free, no API key.

Licence-cleared, quality-scored code components for coding agents (1.4M+ from GitHub). Free search with an API key; every result carries its licence and attribution.

Launch your product on a weekly launch board straight from your coding agent: draft the listing, verify your domain, add the badge and book a week.

Open-source MCP bridge for Nemeton: create, start, snapshot and control native macOS virtual machines from Claude Desktop, Claude Code or Cursor, including UI automation inside the guest. Runs locally on Apple silicon.

Official BugHerd MCP server. Website feedback for coding agents: screenshots, console logs, replay.

Type-safe Python client for Valorant esports data from VLR.gg.

thumbs gives Claude Code or any MCP client an iPhone to see, tap and type on, through one remote URL (https://mcp.usethumbs.com/mcp) with Google sign-in. Early access runs on the iOS Simulator; free, waitlist, allowlist-only for now.