Development Choices
What it costs to build with agents, MCP servers and no-code automation — each option priced, its limits named, and the conditions that decide between them.
Description
Every page here covers one decision a development team actually has to make. The options are named, what each one costs is stated in the units you get billed in, and the conditions under which it breaks are written down rather than implied.
Pages are signed, dated, and cite their sources, so you can see when something was last checked instead of guessing how old the advice is.
Sections
- AI Agents69 pages
Everything related to agentic flows, harnesses, development with agents, building agents.
Bind agent outputs to their bytes, record how each was produced, and verify the digest, builder, and inputs against policy before use.
How to bound agent work at admission, track queue age, choose overflow behavior, preserve fairness, and return useful overload responses.
Separate caller cancellation from deadlines, propagate both, and reserve part of every agent run budget for cleanup and reporting.
Set separate retention and deletion rules for agent conversations, traces, tool payloads, embeddings, caches, and evaluation data.
Compare developer machines, team-operated servers, and managed services by state, identity, isolation, cost, reproducibility, and blast radius.
Limit agent reads, isolate enforcement, treat build-file writes as execution, gate deletion, and pair disk controls with network egress policy.
- No-Code64 pages
No-code or low-code tools, solutions, libraries, styles, automations
Keep editor labels flexible while public identifiers, overwrite rules, folders, and external keys stay predictable.
How no-code media delivery negotiates image formats, adjusts compression, preserves source assets, and keeps CDN cache variants compatible.
Build a no-code removal flow that handles pending jobs, preserves transparency, reviews difficult edges, and publishes only approved assets.
Build mutually exclusive MediaFlows branches, normalize metadata before testing it, and route unknown values through an observable final else path.
Queue triggers, serialize by asset or record, cap shared dependencies, and test backlog behavior without forcing unrelated work into one lane.
Choose the right Power Platform binding for connector credentials, target configuration, automated imports, and secrets.
- MCP67 pages
Everything relevant to MCP servers: installation, usage, custom SKILL files, optimization, development with MCP servers
Compare URL delivery, signed uploads, vendor skills and MCP servers, then add live access only when the agent’s task requires it.
Cloudinary's Asset Management MCP server: asset CRUD, search expressions, transformation URLs, folders, tags and deletion controls.
Separate missed calls, server rejections, API failures, rate limits, and truncated results, then reproduce the exact request through REST.
List an MCP server's resources, select the URI that belongs in context, read it, and handle text or binary content by MIME type.
What the Environment Config MCP server changes, why shared presets and transformations have broad reach, and what reviewers need to record.
How MCP 2026-07-28 removes handshake and session state, makes requests self-describing, and keeps application state explicit.