10 posts on building AI agents with MCP — hosted tools, agent memory, x402 payments, executing templates, and deep-research workflows.
MCP is a small JSON-RPC protocol for connecting a model to tools. Here's what it actually specifies, and the concrete trade-offs of running your own server versus connecting to…
Namespaces, TTLs, versioned history, ACL sharing, and three different ways to search it back out — a tour of WingMan's memory primitives and the patterns that make them useful…
x402 turns HTTP's dormant 402 status code into a real micropayment handshake: agents sign a USDC authorization on Base and retry, with no account, invoice, or human in the loop.…
A tool count is a claim, not a decoration. Here's the actual machinery — an import-time assertion, a live-sync script, and a publish gate — that keeps WingMan's numbers matching…
Most agent runs vanish when the process exits. Here's how to give an agent a durable handle, memory, and a one-call recovery path using WingMan's real MCP tools.
WingMan's 271 hosted tools and its discovery directory are two different things solving two different problems. A tour of both — mcp servers, skills, agents, ai tools, workflows…
Concrete, rebar, framing, joists, board feet, and a bid that ties them together — walking through two of WingMan's trades calculators end to end, with the actual chained template…
Context is the scarcest resource an agent has mid-turn. Errands and watches move the slow, token-heavy work outside it entirely — submit a job, get a job_id back immediately,…
Research, write, and host a citeable report without ever pulling the work into your own context window — chaining submit_errand, check_errand, and store_artifact into one workflow.
Most template galleries show a card and a 'launching soon' badge. WingMan's 15 agent templates are parameterized tool chains that run live, threading each step's output into the…