AllMCPs Blog & Guides
In-depth tutorials, protocol deep dives, and production architecture notes for Model Context Protocol builders.
33 articles
LatestMCP Server Stateless Scaling: Sessions and Routing
Learn how MCP server stateless scaling works with Streamable HTTP, including session IDs, protocol headers, load-balancer routing, external state, and serverless limits.
Read articleMCP Roots: How Clients Scope Server Filesystem Access
MCP roots let clients announce approved filesystem locations to servers. Learn the roots capability, roots/list requests, change notifications, and safe boundaries.
MCP Servers for Data Analysis: Database to Chart
Learn how MCP servers for data analysis connect a database, file server, and code sandbox to turn a question into a validated chart.
MCP Server Docker Image: Run Stdio Servers with Docker
Learn how to build and run an MCP server Docker image for stdio clients, including docker run -i, volume mounts, secrets, debugging, and distribution trade-offs.
Use MCP Servers With a Local LLM: Ollama, Clients, and Tools
Learn how to use MCP servers with a local LLM using Ollama, LM Studio, Cline, and Continue, with setup steps, tool-calling tests, and limits.
Wrap REST API as MCP Server: Task-Shaped Tool Design
Learn how to wrap a REST API as an MCP server by selecting useful endpoints, designing task-shaped tools, forwarding auth, and handling pagination.
MCP Structured Tool Output: outputSchema and structuredContent
Learn when MCP tools should return plain text or structuredContent with an outputSchema, how protocol versions affect support, and how clients render each.
MCP Server Error Handling: Protocol vs Tool Errors
Learn MCP server error handling: distinguish JSON-RPC protocol failures from isError tool results, and return messages models can understand and recover from.
MCP OAuth Authorization: Step-by-Step PKCE and Tokens
Learn the MCP OAuth authorization flow step by step, including protected resource metadata, dynamic client registration, PKCE, and token audience validation.
Build an MCP Server with the TypeScript SDK
Build an MCP server with the TypeScript SDK from zero: register a tool and resource, test both with an in-memory client, and compare stdio with HTTP.
Build an MCP Server in Python with FastMCP and uv
Build an MCP server in Python with FastMCP using decorators, typed tool arguments, context logging, and uv packaging that runs through uvx.
MCP Tool Annotations: readOnlyHint, destructiveHint, and What Clients Do With Them
Tool annotations are four booleans and a title that tell a client whether your tool reads, writes, deletes, or reaches the outside world. Here is what each hint means, the defaults that bite servers that skip them, and why clients must treat every hint as untrusted.
Migrating MCP Servers from HTTP+SSE to Streamable HTTP
The HTTP+SSE transport is deprecated and the dual-endpoint session model is the reason. Here is what Streamable HTTP actually changes, how to run both transports from one handler during the window, and the proxy and infrastructure settings that quietly break the migration.
Multi Round-Trip Requests: Porting MCP Servers Off Server-Initiated Calls
The 2026-07-28 revision removed a server's ability to send its own JSON-RPC request mid-tool-call. MRTR replaces sampling, elicitation, and roots with a resumable input_required result. Here is how to port a real server, what requestState has to carry, and how to sign it so the round trip is safe.
MCP Elicitation: How Servers Ask the User for Input Mid-Task
Sampling lets an MCP server ask the client for LLM reasoning. Elicitation is the other half — a server pausing mid-tool-call to ask the human for a missing value, a confirmation, or a choice. Here is how elicitation/create works, with current TypeScript and Python code.
MCP Progress Notifications and Subscriptions: Handling Long-Running Tasks and Live Updates
Long-running MCP tools often fail silently or hit hard client timeouts in Claude and Cursor. Here is how to implement JSON-RPC progress tokens, resource subscriptions, logging levels, and cancellation in TypeScript and Python.
MCP Sampling Explained: How to Build AI-Powered MCP Servers Without API Keys
MCP servers don't have to be dumb pipes. With MCP sampling, your server can request LLM reasoning directly through the host client — without bundling API keys, managing token billing, or leaking secrets. Here is how sampling works with TypeScript and Python implementations.
MCP Resources and Prompts: The Primitives Most Servers Skip
Tools get all the attention, but MCP defines two other primitives — resources and prompts — with a completely different control model. Here's what they actually do, real TypeScript and Python code, and why most servers never implement them.
Claude Skills vs MCP: What's the Difference and When to Use Each
Skills and MCP servers both extend what Claude can do, and both ship from Anthropic, which is exactly why they get conflated. Here's the actual distinction — procedural knowledge vs live connections — and a practical framework for which one a given problem needs.
MCP Registry vs MCP Directory: What the Official Registry Is and How to Publish to It
The official MCP registry, third-party directories, and in-client marketplaces are three different things doing three different jobs. Here's how they fit together, how to publish your server to the official registry with server.json, and where a curated directory listing still matters.
How Many MCP Servers Is Too Many? Tool Overload and Context Budgets
Every MCP server you connect adds its full tool list to every request, whether you use it or not. A practical guide to measuring the cost, spotting overload, and trimming your setup without losing capability.
Browser Automation MCP Servers: Playwright vs Puppeteer vs Browser-Use vs Selenium
A practical decision framework for picking a browser automation MCP server — what Playwright, Puppeteer, browser-use, and Selenium wrappers actually differ on, plus the headless, auth, and security tradeoffs that matter once an agent is driving a real browser.
Writing MCP Tool Descriptions That Actually Get Picked
The model never reads your code — only your tool's name, description, and schema. A practical guide to writing MCP tool descriptions that get selected correctly, with the arguments filled in right, every time.
Testing and Debugging MCP Servers: From Inspector to CI
A practical guide to testing Model Context Protocol servers before you ship — the Inspector, unit-testing tool handlers, integration harnesses, structured logging, and a pre-publish checklist.
Deploying Remote MCP Servers: The Complete Production & Cloud Hosting Blueprint
A complete guide to deploying remote Model Context Protocol (MCP) servers to production on Cloudflare Workers, Docker, Fly.io, and AWS with SSE transports, CORS, SSL, and secrets management.
What Can MCP Servers Do? A Field Guide to the Model Context Protocol Ecosystem
What MCP servers actually do, how the 3,000+ available servers break down by category, and how to choose the right Model Context Protocol server for your AI agent.
Securing & Authenticating Remote MCP Servers: The Production Guide
A complete guide to authenticating and securing remote Model Context Protocol (MCP) servers over HTTP and SSE, featuring OAuth 2.0 PKCE, JWT bearer validation, multi-tenant token propagation, and prompt injection defense.
How to get a free dofollow backlink for your MCP server
Claim your listing, verify your product site, and place a dofollow AllMCPs badge — reciprocal SEO that helps your product and the directory.
How to compare MCP servers (and pick the right alternative)
A practical framework for evaluating Model Context Protocol servers — tools, install path, health, and side-by-side comparisons on AllMCPs.
How to Install an MCP Server in Claude, Cursor, Windsurf, and VS Code
One reference guide to installing Model Context Protocol servers across every major client — the universal config shape, exact file locations, and how to verify your tools loaded.
MCP Server Not Connecting? A Troubleshooting Guide for Every Client
The most common reasons an MCP server fails to connect, shows zero tools, or crashes on startup — and exactly how to fix each one, across Claude Desktop, Claude Code, Cursor, and other MCP clients.
Architecting Production-Ready MCP Servers: Security, Transports, and Tool Design Patterns
A practical engineering guide to building secure, performant Model Context Protocol (MCP) servers with robust transports, token-optimized schemas, and defense-in-depth authorization.
Launching AllMCPs: A Directory for Model Context Protocol Servers
Why MCP server discovery is broken, how AllMCPs fixes it, and a practical guide to evaluating, installing, and claiming servers in the directory.