All 44 articles, sorted alphabetically
MCP authorization architecture
Deep-dive on MCP's authorization spec: protected-resource metadata discovery, dynamic client registration, PKCE consent flows, audience-bound acc…
Read article →MCP Authentication Models, in depth: stdio credentials, user-delegated OAuth, client credentials, enterprise-managed ID-JAG, and how to choose
A decision guide to authenticating Model Context Protocol clients and servers: environment credentials for stdio, static keys and why they sit outside…
Read article →MCP request cancellation architecture
How cancellation works in the Model Context Protocol: the notifications/cancelled message and why it is fire-and-forget, the unavoidable race between …
Read article →MCP capability negotiation architecture
Deep-dive on MCP capability negotiation: the initialize handshake where client and server exchange protocol versions and capabilities, the intersectio…
Read article →MCP Client Architecture in Depth: stateless requests, era detection, catalogs, MRTR and subscriptions
Build an MCP client layer for the 2026-07-28 stateless protocol and legacy servers: per-request _meta, era probes, namespaced catalogs, the input_requ…
Read article →MCP Client and Server
The MCP role model in detail: why there are three roles and not two, what a host owns versus a client versus a server, the one-client-per-server relat…
Read article →MCP completion -- argument autocompletion for tools and prompts
Deep-dive on the MCP completion capability: the argument-filling need, the server offering autocompletion suggestions, context-awareness, completion f…
Read article →MCP elicitation
How MCP elicitation lets a server ask the user for input mid-operation: why a server cannot prompt directly, the schema that constrains what may be as…
Read article →MCP error handling architecture
Deep-dive on Model Context Protocol error handling: the JSON-RPC code space, why tool failures ride inside successful results with isError, cancellati…
Read article →MCP gateway architecture
Deep-dive on MCP gateways: merged policy-filtered catalogs, namespace mapping, OAuth token exchange per backend, session and notification fan-out, tra…
Read article →MCP logging architecture
How MCP server logging works: the logging capability and logging/setLevel, the notifications/message shape with level, logger and data, the eight RFC …
Read article →Multi-tenant MCP server architecture
Deep-dive on multi-tenancy for MCP servers: why isolation must live in the authenticated session rather than tool arguments, how identity flows from t…
Read article →MCP Notifications: Fire-and-Forget Messaging in the Model Context Protocol
How MCP uses id-less JSON-RPC notifications for list-changed events, progress, resource updates, and cancellation — and the idempotent, backpressure-a…
Read article →Model Context Protocol Overview
A map of the Model Context Protocol: the N x M integration problem it solves, what MCP is as an open JSON-RPC protocol, the host, client and server ro…
Read article →MCP pagination architecture
Deep-dive on Model Context Protocol pagination: cursor-based paging for resources/tools/prompts lists, opaque nextCursor tokens, keyset vs offset resu…
Read article →MCP progress architecture
How MCP reports progress on long-running operations: the progress token a client attaches to opt in, the fields a progress notification carries, why p…
Read article →MCP Prompt Templates, in depth: engineering the server side, from argument parsing and safe rendering to embedded resources, a dynamic registry and golden tests
How to build MCP prompt templates that hold up in production: string arguments parsed and validated, rendering that keeps untrusted text out of the in…
Read article →MCP prompts architecture
MCP prompts explained as the user-invoked primitive: prompts/list and prompts/get, declaring arguments, why a prompt returns a message sequence instea…
Read article →MCP Protocol Details
How MCP messages actually look on the wire: JSON-RPC 2.0 as the substrate, the three message kinds, request IDs and correlation, the method namespace,…
Read article →MCP rate limiting architecture
Deep-dive on rate limiting an MCP tool server: how a per-client, per-tool token bucket admits or rejects each call — weighting expensive tools more, a…
Read article →MCP server registry -- discovering and managing servers
Deep-dive on MCP server registries: the manual-config scaling problem, the registry catalog, capability-based discovery, server metadata, publishing, …
Read article →Remote MCP Servers in 2026, in depth: the stateless Streamable HTTP revision, mirrored headers, MRTR, authorization, and serving older clients
How to build and operate a remote Model Context Protocol server against the 2026-07-28 revision: per-request metadata instead of sessions, the single …
Read article →MCP resource subscriptions architecture
Deep-dive on MCP resource subscriptions: capability negotiation, resources/subscribe, the resources/updated notification, server-side subscription reg…
Read article →MCP resources architecture
How MCP resources work: the application-controlled context primitive, why a resource is not a tool, URI addressing and URI templates, resources/list v…
Read article →MCP roots architecture
How MCP roots work: the client declaring which URIs a server may operate on, the roots/list request and the list-changed notification, why a root is a…
Read article →MCP sampling architecture
How MCP sampling inverts the usual call: an MCP server asks the host to run an LLM completion on its behalf. The request shape, model preferences as h…
Read article →MCP Security, in depth: a threat model for clients, servers and proxies, from SSRF in OAuth discovery to tool poisoning
A threat-by-threat guide to securing Model Context Protocol deployments: trust boundaries, confused deputy and token passthrough, SSRF during OAuth di…
Read article →MCP Security Model, in depth: roles, trust boundaries and enforcement points in the stateless 2026-07-28 protocol
The MCP security model as of the 2026-07-28 revision: who is trusted with what, the three spec principles, per-request authorization, sealed requestSt…
Read article →MCP Server Architecture in Depth
MCP server architecture in depth for the stateless 2026-07-28 protocol: request path, envelope validation code, tools and resources, multi round-trip …
Read article →MCP Server Design Patterns, in depth: facades, primitives, error contracts, state, long-running work and composition
A pattern catalogue for designing Model Context Protocol servers against the 2025-11-25 specification: API mirror vs workflow facade, choosing tools, …
Read article →MCP sessions -- stateful connections between client and server
Deep-dive on MCP sessions: the stateful client-server connection, initialization (handshake, version), capability negotiation (extensibility and compa…
Read article →MCP SSE Transport, in depth: the deprecated HTTP+SSE wire protocol, running it behind proxies, and migrating to Streamable HTTP
A practical guide to the Model Context Protocol HTTP+SSE transport from protocol version 2024-11-05: the two endpoints and the endpoint event, SSE fra…
Read article →MCP stdio Transport, in depth: spawning servers, line framing, stderr, shutdown and debugging a local MCP server
How the Model Context Protocol stdio transport works at the process level: what the 2025-11-25 specification requires, launching a server with its env…
Read article →MCP structured tool output architecture
A focused guide to structured tool output in the Model Context Protocol: why prose-only tool results force the host to re-parse text and the model to …
Read article →MCP Server Testing Patterns, in depth: in-memory clients, contract snapshots, error paths and model evals
How to test an MCP server properly: a five-layer test pyramid, unit tests of tool logic, in-memory protocol tests with the Python SDK's Client, s…
Read article →MCP tool annotations architecture
Deep-dive on MCP tool annotations: readOnlyHint, destructiveHint, idempotentHint, openWorldHint, and title; how clients turn them into auto-run/confir…
Read article →MCP Tools Architecture in Depth
How MCP tools actually work: tools/list and tools/call, the tool definition of name, description and input schema, why the description is prompt surfa…
Read article →MCP transport architecture
MCP transport as an abstraction: what the protocol requires of any wire (bidirectional messages, framing, ordering, correlation), why one JSON-RPC lay…
Read article →MCP Transport Layer Deep-Dive: stdio vs. SSE for Remote Resources
The Model Context Protocol (MCP) provides a powerful, standardized contract for how an AI agent invokes a tool.
Read article →MCP streaming transport architecture
Deep-dive on MCP streaming transport: SSE frames, chunked deltas, progress events, backpressure, cancel, reconnect.
Read article →MCP versioning architecture
Deep-dive on Model Context Protocol versioning: how the initialize handshake negotiates a shared protocol version, how capability flags let features e…
Read article →MCP Versioning Strategy, in depth: modern and legacy eras, per-request versions, dual-era servers, deprecation windows and versioning your own tools
A practical versioning strategy for teams that build or operate Model Context Protocol servers and clients: the three things that version independentl…
Read article →MCP vs OpenAI Function Calling, in depth: two layers, an adapter between them, strict schemas and the hosted MCP tool
Why MCP and OpenAI function calling are not competitors: function calling is the contract between an application and the model, MCP is the protocol be…
Read article →Solving the "N x M" Integration Problem with MCP Servers and Clients
In modern enterprises, the AI landscape is exploding. A company may have N distinct AI models and agents—a summarization agent, a customer support chatbot, a data…
Read article →