MCP
Everything relevant to MCP servers: installation, usage, custom SKILL files, optimization, development with MCP servers
How an IDE assistant and a desktop assistant differ as MCP clients: config shape, repository context, tool approval, OAuth reach, and marketplace plugins.
OAuth binds an MCP server connection to one product environment with no secret on disk. Header auth works headless and holds several environments at once.
REST is deterministic code you can test and review; an MCP server lets a model decide per item. How cost, failure, auditability split and which to build first.
How a remote MCP server changes underneath a client, why versioned paths and deprecated transports matter, and how to assert the tool surface at session start.
How to add retries to agent-driven MCP tool calls without amplifying rate limits: classify errors, back off with jitter, keep retries out of the model.
What to log, what to keep, and how to tie an agent's MCP tool calls to a vendor's own records so a surprising asset change can be explained days later.
How the cloudinary-region header selects the API region an MCP server talks to, why a wrong value fails silently, and what it does not do for data residency.
How to move an MCP-driven step from a developer machine into CI: header credentials, secret store, tool allowlist, outcome assertions and shared rate limits.
Remote and local MCP servers expose the same tools. Compare runtime cost, credential handling, update ownership, and the one condition that settles the choice.
Cloudinary's remote MCP servers use stateless Streamable HTTP at /mcp; /sse is deprecated except for Analysis, and still accepts POST as an alias.
Why a listing call over a large Cloudinary library can silently truncate inside an MCP client, how to confirm it, and how to ask for less instead.
What the machine-readable transformation rules file is, how MCP transformation tools use it, and what it does and does not tell you about a URL.
Step-by-step setup for Cloudinary's MediaFlows MCP server: the /v2/mcp endpoint, its three custom auth headers, and why its config differs from the others.
How to run upload moderation from an agent through Cloudinary's Analysis MCP server: two thresholds, held delivery, human review, sampling and a decision log.
How to create and update Cloudinary named transformations through an MCP server, and why the update step is both the point and the risk.
How to turn on cloudinary-embed-headers so every MCP tool result carries the rate-limit ceiling, remaining allowance, reset time and request ID.
Use the cloudinary-tools header to allowlist tools on a remote MCP connection: fewer definitions in context, better targeting, and server-side enforcement.
What Cloudinary's Structured Metadata MCP server does: typed fields, controlled values, conditional rules, stable external IDs, and search-expression queries.
A skill is instruction text; an MCP server executes against a live account. How Cloudinary's four-skill pack and its MCP servers differ, and when to use each.