WEBVTT - Agents shouldn't need to scrape your blog 1 00:00:00.000 --> 00:00:03.264 Agents shouldn't need to scrape your blog 2 00:00:03.264 --> 00:00:06.527 I shipped two MCP tools to mondello. 3 00:00:06.527 --> 00:00:09.325 dev this morning: list_skills and get_skill_preview. 4 00:00:09.325 --> 00:00:15.386 The skills marketplace I launched three days ago already had four agent-facing surfaces. 5 00:00:15.386 --> 00:00:17.717 /skills is the HTML storefront. 6 00:00:17.717 --> 00:00:18.183 /skills. 7 00:00:18.183 --> 00:00:20.514 json is a machine-readable manifest. 8 00:00:20.514 --> 00:00:22.845 /skills/<slug>/raw is the x402-gated download. 9 00:00:22.845 --> 00:00:25.176 /skills/<slug>/install is the bash installer. 10 00:00:25.176 --> 00:00:27.974 What it didn't have was MCP. 11 00:00:27.974 --> 00:00:40.562 Without MCP, an agent installing a skill had to either fetch and parse a JSON endpoint it had to already know about, or scrape the HTML storefront. 12 00:00:40.562 --> 00:00:41.494 Both work. 13 00:00:41.494 --> 00:00:44.758 Neither is how agents should discover anything. 14 00:00:44.758 --> 00:00:46.157 What was broken 15 00:00:46.157 --> 00:00:50.819 The /skills page read, in its "how installation works" section: 16 00:00:50.819 --> 00:01:00.144 Any MCP agent that speaks x402 can pay directly against /skills/<slug>/raw. See /skills.json for the full catalog with payment metadata. 17 00:01:00.144 --> 00:01:02.008 That's half the story. 18 00:01:02.008 --> 00:01:06.205 An agent connected over MCP has a typed protocol. 19 00:01:06.205 --> 00:01:09.468 It shouldn't need to know that /skills. 20 00:01:09.468 --> 00:01:10.401 json exists. 21 00:01:10.401 --> 00:01:19.259 It shouldn't need to fetch an HTTP URL, parse JSON, map fields, and then compose its own install command. 22 00:01:19.259 --> 00:01:24.854 It should ask the MCP server for the catalog and the prices. 23 00:01:24.854 --> 00:01:26.719 The server should answer. 24 00:01:26.719 --> 00:01:29.982 Before this morning, that conversation couldn't happen. 25 00:01:29.982 --> 00:01:39.773 /api/mcp exposed fourteen tools — posts, cues, changelog entries, site stats — and zero tools about the marketplace one directory away. 26 00:01:39.773 --> 00:01:42.104 The other fourteen didn't fit. 27 00:01:42.104 --> 00:01:43.503 Skills aren't posts. 28 00:01:43.503 --> 00:01:50.496 They have prices, payment metadata, and a paid/preview split that list_posts and get_post_markdown don't model. 29 00:01:50.496 --> 00:01:51.895 What I shipped 30 00:01:51.895 --> 00:01:56.091 Two tools. About 165 lines of TypeScript in src/pages/api/mcp.ts. 31 00:01:56.091 --> 00:01:58.422 list_skills returns the full catalog. 32 00:01:58.422 --> 00:02:03.551 Every published skill, with title, excerpt, price, and the install one-liner. 33 00:02:03.551 --> 00:02:14.274 It also returns currency: "USDC" and network: "eip155:8453" at the top level so an x402-aware agent can budget the round-trip before fetching anything: 34 00:02:14.274 --> 00:02:18.936 get_skill_preview returns the free preview for one skill by slug. 35 00:02:18.936 --> 00:02:25.463 Not the body — the body is behind x402 at /raw and stays there. 36 00:02:25.463 --> 00:02:37.119 The preview comes from a dedicated preview portableText field I author in the admin UI alongside the paid body, with the excerpt as a fallback. 37 00:02:37.119 --> 00:02:47.376 A few paragraphs that demonstrate the shape of the skill so an agent (or a human) can decide if it's worth buying. 38 00:02:47.376 --> 00:03:01.363 Both wire into the existing JSON-RPC 2.0 dispatcher alongside search_cues, list_posts, get_post_markdown, and the rest. No special-casing, no new auth, no new routes. Two more entries in the tool catalog. 39 00:03:01.363 --> 00:03:03.694 The walk, end to end 40 00:03:03.694 --> 00:03:10.221 Here's what an MCP-connected agent does now to discover, preview, and install a skill: 41 00:03:10.221 --> 00:03:10.688 1. 42 00:03:10.688 --> 00:03:16.749 tools/list — sees list_skills and get_skill_preview in the catalog alongside the content tools. 43 00:03:16.749 --> 00:03:17.215 2. 44 00:03:17.215 --> 00:03:25.607 list_skills — gets the eight skills, picks the one that matches the task at hand, notes the price. 45 00:03:25.607 --> 00:03:26.073 3. 46 00:03:26.073 --> 00:03:30.735 get_skill_preview(slug) — reads the preview, decides the skill is relevant. 47 00:03:30.735 --> 00:03:31.202 4. 48 00:03:31.202 --> 00:03:35.398 GET /skills/<slug>/raw with x402 payment header — settles $0. 49 00:03:35.398 --> 00:03:38.661 10 USDC on Base, receives the markdown. 50 00:03:38.661 --> 00:03:39.128 5. 51 00:03:39.128 --> 00:03:40.526 Writes to ~/. 52 00:03:40.526 --> 00:03:40.992 claude/skills/<slug>/skill. 53 00:03:40.992 --> 00:03:46.587 md and the skill is live for the rest of the session. 54 00:03:46.587 --> 00:03:47.986 Five HTTP requests. 55 00:03:47.986 --> 00:03:48.918 One payment. 56 00:03:48.918 --> 00:03:50.783 No HTML in sight. 57 00:03:50.783 --> 00:03:59.642 The win isn't raw latency — a single curl of a well-structured index beats five round-trips on the stopwatch. 58 00:03:59.642 --> 00:04:01.507 The win is stability. 59 00:04:01.507 --> 00:04:06.169 Every step returns a typed payload with a stable schema. 60 00:04:06.169 --> 00:04:07.101 Nothing renders. 61 00:04:07.101 --> 00:04:08.034 Nothing lazy-loads. 62 00:04:08.034 --> 00:04:12.696 The CSS can change tomorrow and this flow is unaffected. 63 00:04:12.696 --> 00:04:14.095 Why this matters 64 00:04:14.095 --> 00:04:21.088 Yes, "build an API, not a scraper" is a fifteen-year-old argument with a new acronym. 65 00:04:21.088 --> 00:04:23.419 What's new isn't the principle. 66 00:04:23.419 --> 00:04:37.406 It's that MCP is a single client-side protocol the biggest agent clients already speak, so "ship an API" no longer means "and convince every consumer to integrate with it separately." 67 00:04:37.406 --> 00:04:39.737 That asymmetry is the unlock. 68 00:04:39.737 --> 00:04:44.866 Most sites that say "agents welcome" mean "we won't block you." 69 00:04:44.866 --> 00:04:48.596 They ship HTML and hope the scraper holds. 70 00:04:48.596 --> 00:04:55.589 It holds until the layout shifts, an ad renders mid-content, or a component loads async. 71 00:04:55.589 --> 00:05:02.583 Well-structured semantic HTML is a real contract — I'm not pretending otherwise — and /skills. 72 00:05:02.583 --> 00:05:10.975 json is a legitimate middle ground for any agent that'd rather fetch a typed manifest than speak RPC. 73 00:05:10.975 --> 00:05:18.901 But scrapers are still a hack we tolerate because writing an API felt like too much work. 74 00:05:18.901 --> 00:05:21.232 MCP makes the alternative easy. 75 00:05:21.232 --> 00:05:23.097 The protocol is small. 76 00:05:23.097 --> 00:05:24.962 The transport is HTTP. 77 00:05:24.962 --> 00:05:30.556 The tool shape is a name, a description, and a JSON schema. 78 00:05:30.556 --> 00:05:37.083 If a site already has an MCP server, a new tool is fifty lines. 79 00:05:37.083 --> 00:05:39.415 An agent interacting with mondello. 80 00:05:39.415 --> 00:05:43.611 dev doesn't need to know what a blog is. 81 00:05:43.611 --> 00:05:47.807 It doesn't need to know what HTML looks like. 82 00:05:47.807 --> 00:05:54.334 It needs to know MCP and x402, both documented in under ten pages each. 83 00:05:54.334 --> 00:06:06.922 Everything else is the site telling the agent, in the protocols the agent already understands, exactly what's for sale, what it costs, and how to buy it. 84 00:06:06.922 --> 00:06:09.253 The missing piece is discovery. 85 00:06:09.253 --> 00:06:14.848 An agent only calls list_skills if someone first registers the MCP server. 86 00:06:14.848 --> 00:06:15.314 /. 87 00:06:15.314 --> 00:06:21.375 well-known/mcp and agent-side registries are what solves that, and neither is solved today. 88 00:06:21.375 --> 00:06:30.234 Until they are, MCP discovery is an out-of-band step — somebody has to tell the agent where to look. 89 00:06:30.234 --> 00:06:32.098 That's the live problem. 90 00:06:32.098 --> 00:06:33.497 The bigger pattern 91 00:06:33.497 --> 00:06:36.295 Same pattern, everywhere on this blog: 92 00:06:36.295 --> 00:06:42.356 Posts are served as Markdown with YAML frontmatter at /<slug>.md, not just HTML. 93 00:06:42.356 --> 00:06:52.146 Search results come back with audio-timestamped deep-links in /api/search/cues.json, not embedded in a page that you have to render and click. 94 00:06:52.146 --> 00:07:00.072 The full archive is one MCP call (get_archive) away, returning every post as one concatenated Markdown document. 95 00:07:00.072 --> 00:07:10.329 Even the visual stuff — the OG cards with per-post audio waveforms, the social share previews — are SVG files served directly. 96 00:07:10.329 --> 00:07:13.593 The blog is the UI for humans. 97 00:07:13.593 --> 00:07:16.856 The protocols are the UI for agents. 98 00:07:16.856 --> 00:07:20.120 Same content, same database, same authoring workflow. 99 00:07:20.120 --> 00:07:21.053 Nothing duplicated. 100 00:07:21.053 --> 00:07:22.917 Nothing "adapted for AI." 101 00:07:22.917 --> 00:07:24.316 Posts are posts. 102 00:07:24.316 --> 00:07:25.715 Skills are skills. 103 00:07:25.715 --> 00:07:33.641 MCP and x402 just mean both audiences can consume them without one pretending to be the other. 104 00:07:33.641 --> 00:07:38.303 If you want to see it from the agent side: 105 00:07:38.303 --> 00:07:50.891 Register the server, ask Claude for the skills catalog, and it has a typed tool that fits — it'll reach for list_skills before it reaches for fetch. 106 00:07:50.891 --> 00:07:54.621 Five round-trips, one payment, no HTML in sight. 107 00:07:54.621 --> 00:07:57.418 The discovery is still on you. 108 00:07:57.418 --> 00:07:58.817 The rest isn't. 109 00:07:58.817 --> 00:08:09.074 The build-and-ship ritual that deploys this post — Astro build, post-build hook patching, wrangler deploy — is itself a skill at /skills/emdash-deploy. 110 00:08:09.074 --> 00:08:15.135 Same pattern: one curl command, x402 USDC, the recipe lives in your ~/. 111 00:08:15.135 --> 00:08:17.000 claude/skills/ directory after payment.