read-only demo

Videos sAOBXCDiDOs

MCP Apps: Primitives, discovery, and the Future of Software - Pietro Zullo, Manufact, Inc

index_state ready data_status ok

AI Engineer· published 2026-07-05· 0:28:54· en-US· indexed 2026-08-11 05:23

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:33, 1 of 1 keyframes kept
  2. Shot 1, 0:33 to 1:05, 1 of 1 keyframes kept
  3. Shot 2, 1:05 to 1:37, 0 of 1 keyframes kept
  4. Shot 3, 1:37 to 2:05, 1 of 1 keyframes kept
  5. Shot 4, 2:05 to 2:34, 0 of 1 keyframes kept
  6. Shot 5, 2:34 to 3:02, 0 of 1 keyframes kept
  7. Shot 6, 3:02 to 3:31, 1 of 1 keyframes kept
  8. Shot 7, 3:31 to 4:01, 0 of 1 keyframes kept
  9. Shot 8, 4:01 to 4:30, 0 of 1 keyframes kept
  10. Shot 9, 4:30 to 5:00, 0 of 1 keyframes kept
  11. Shot 10, 5:00 to 5:29, 0 of 1 keyframes kept
  12. Shot 11, 5:29 to 6:01, 1 of 1 keyframes kept
  13. Shot 12, 6:01 to 6:08, 1 of 1 keyframes kept
  14. Shot 13, 6:08 to 6:50, 1 of 1 keyframes kept
  15. Shot 14, 6:50 to 7:30, 1 of 1 keyframes kept
  16. Shot 15, 7:30 to 8:12, 1 of 1 keyframes kept
  17. Shot 16, 8:12 to 8:15, 1 of 1 keyframes kept
  18. Shot 17, 8:15 to 8:41, 1 of 1 keyframes kept
  19. Shot 18, 8:41 to 9:07, 0 of 1 keyframes kept
  20. Shot 19, 9:07 to 9:33, 0 of 1 keyframes kept
  21. Shot 20, 9:33 to 9:59, 0 of 1 keyframes kept
  22. Shot 21, 9:59 to 10:25, 1 of 1 keyframes kept
  23. Shot 22, 10:25 to 10:50, 1 of 1 keyframes kept
  24. Shot 23, 10:50 to 11:15, 0 of 1 keyframes kept
  25. Shot 24, 11:15 to 11:50, 1 of 1 keyframes kept
  26. Shot 25, 11:50 to 12:25, 1 of 1 keyframes kept
  27. Shot 26, 12:25 to 13:01, 1 of 1 keyframes kept
  28. Shot 27, 13:01 to 13:29, 1 of 1 keyframes kept
  29. Shot 28, 13:29 to 13:57, 0 of 1 keyframes kept
  30. Shot 29, 13:57 to 14:25, 0 of 1 keyframes kept
  31. Shot 30, 14:25 to 14:53, 1 of 1 keyframes kept
  32. Shot 31, 14:53 to 15:21, 0 of 1 keyframes kept
  33. Shot 32, 15:21 to 15:57, 1 of 1 keyframes kept
  34. Shot 33, 15:57 to 16:34, 0 of 1 keyframes kept
  35. Shot 34, 16:34 to 16:38, 1 of 1 keyframes kept
  36. Shot 35, 16:38 to 16:52, 1 of 1 keyframes kept
  37. Shot 36, 16:52 to 17:00, 1 of 1 keyframes kept
  38. Shot 37, 17:00 to 17:01, 1 of 1 keyframes kept
  39. Shot 38, 17:01 to 17:04, 1 of 1 keyframes kept
  40. Shot 39, 17:04 to 17:07, 1 of 1 keyframes kept
  41. Shot 40, 17:07 to 17:08, 1 of 1 keyframes kept
  42. Shot 41, 17:08 to 17:15, 1 of 1 keyframes kept
  43. Shot 42, 17:15 to 17:18, 1 of 1 keyframes kept
  44. Shot 43, 17:18 to 17:19, 0 of 1 keyframes kept
  45. Shot 44, 17:19 to 17:20, 0 of 1 keyframes kept
  46. Shot 45, 17:20 to 17:23, 1 of 1 keyframes kept
  47. Shot 46, 17:23 to 17:48, 1 of 1 keyframes kept
  48. Shot 47, 17:48 to 18:11, 1 of 1 keyframes kept
  49. Shot 48, 18:11 to 18:13, 1 of 1 keyframes kept
  50. Shot 49, 18:13 to 18:35, 1 of 1 keyframes kept
  51. Shot 50, 18:35 to 19:02, 1 of 1 keyframes kept
  52. Shot 51, 19:02 to 19:29, 1 of 1 keyframes kept
  53. Shot 52, 19:29 to 19:56, 1 of 1 keyframes kept
  54. Shot 53, 19:56 to 20:23, 1 of 1 keyframes kept
  55. Shot 54, 20:23 to 20:27, 1 of 1 keyframes kept
  56. Shot 55, 20:27 to 20:52, 1 of 1 keyframes kept
  57. Shot 56, 20:52 to 21:40, 0 of 1 keyframes kept
  58. Shot 57, 21:40 to 22:00, 1 of 1 keyframes kept
  59. Shot 58, 22:00 to 22:26, 1 of 1 keyframes kept
  60. Shot 59, 22:26 to 22:52, 0 of 1 keyframes kept
  61. Shot 60, 22:52 to 23:19, 0 of 1 keyframes kept
  62. Shot 61, 23:19 to 23:45, 0 of 1 keyframes kept
  63. Shot 62, 23:45 to 24:11, 0 of 1 keyframes kept
  64. Shot 63, 24:11 to 24:38, 0 of 1 keyframes kept
  65. Shot 64, 24:38 to 25:09, 1 of 1 keyframes kept
  66. Shot 65, 25:09 to 25:41, 0 of 1 keyframes kept
  67. Shot 66, 25:41 to 26:13, 0 of 1 keyframes kept
  68. Shot 67, 26:13 to 26:40, 1 of 1 keyframes kept
  69. Shot 68, 26:40 to 27:06, 0 of 1 keyframes kept
  70. Shot 69, 27:06 to 27:33, 0 of 1 keyframes kept
  71. Shot 70, 27:33 to 28:00, 0 of 1 keyframes kept
  72. Shot 71, 28:00 to 28:27, 0 of 1 keyframes kept
  73. Shot 72, 28:27 to 28:53, 1 of 1 keyframes kept

73 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
269
whisperx 269
chunks
51
from 269 cues
keyframes
44
kept of 73 captured
frames with text
44
901 lines read
chapters
0
from the source metadata
keyframe bytes
6.6 MB
word timings on 269 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done 2026-08-11 05:18 1m 20s
stt done 2026-08-11 05:20 29s
chunk done 2026-08-11 05:20 0s
text_embed done 2026-08-11 05:20 0s
keyframe done 2026-08-11 05:20 2m 23s
ocr done 2026-08-11 05:23 20s
frame_embed done 2026-08-11 05:23 7s

Frames, and what the machine read

  • 0:19 #0 done7 line(s)

    shot 0·sharpness 1976.5

    1. Manufact0.95
    2. MCP APPS0.97
    3. Primitives, discovery, and why this1.00
    4. is how all software will be used0.98
    5. Pietro Zullo1.00
    6. MANUFACT (YC S25) . CREATOR OF MCP-USE . 8M+ DOWNLOADS0.95
    7. MANUFACT.COM1.00
  • 1:01 #1 done8 line(s)

    shot 1·sharpness 2730.9

    1. Manufact0.96
    2. INTRO0.94
    3. You know MCP. Do you know MCP Apps?1.00
    4. Everyone in this room knows what MCP is. Far fewer know1.00
    5. what MCP Apps are, how they work, how to build them,0.99
    6. and how to distribute them.1.00
    7. By the end of this talk you will know everything you need to join the race.1.00
    8. MANUFACT.COM1.00
  • 1:30 #2 skipped

    shot 2·duplicate of #1

  • 2:02 #3 done10 line(s)

    shot 3·sharpness 2116.5

    1. Manufact1.00
    2. WHAT WE BUILD0.97
    3. Open-source SDKs and dev tools for MCP1.00
    4. mcp-use to build0.97
    5. Inspector to test0.98
    6. Manufact to ship0.99
    7. 8M+DOWNLOADS0.98
    8. 10kGITHUB STARS0.99
    9. GITHUB.C1.00
    10. MANUFACT.COM1.00
  • 2:30 #4 skipped

    shot 4·duplicate of #3

  • 2:48 #5 skipped

    shot 5·duplicate of #3

  • 3:05 #6 done24 line(s)

    shot 6·sharpness 1719.4

    1. Manufact1.00
    2. TIMELINE1.00
    3. How we got here0.98
    4. NOV 20240.97
    5. MAY 20251.00
    6. 0CT 20250.92
    7. LATE 20250.95
    8. JAN 20260.96
    9. MCP launches0.97
    10. mcp-ui1.00
    11. ChatGPT Apps SDK0.98
    12. The stores open1.00
    13. MCP Apps is official0.99
    14. Models connect to tools and data.0.99
    15. Open-source project returns1.00
    16. OpenAl brings app Uls to ChatGPT,1.00
    17. ChatGPT and Claude start listing1.00
    18. SEP-1865 merges as the first MCP1.00
    19. Results are text only.1.00
    20. interactive HTML over MCP.0.99
    21. built on MCP.1.00
    22. apps and connectors.0.98
    23. extension.1.00
    24. MANUFACT.COM1.00
  • 3:35 #7 skipped

    shot 7·duplicate of #6

  • 4:04 #8 skipped

    shot 8·duplicate of #6

  • 4:34 #9 skipped

    shot 9·duplicate of #6

  • 5:03 #10 skipped

    shot 10·duplicate of #6

  • 5:33 #11 done29 line(s)

    shot 11·sharpness 1653.8

    1. Manufact1.00
    2. ADOPTION1.00
    3. Apps are landing in the stores, fast1.00
    4. ChatGPT apps & Claude connectors over time0.99
    5. CUMULATIVE COUNT0.97
    6. 8001.00
    7. 6001.00
    8. 4001.00
    9. 2001.00
    10. WI0.90
    11. W20.99
    12. W30.96
    13. W41.00
    14. W50.89
    15. W60.95
    16. W70.96
    17. W80.99
    18. W90.78
    19. W100.86
    20. W110.92
    21. W120.98
    22. W130.99
    23. W141.00
    24. W150.99
    25. W160.98
    26. W170.99
    27. ChatGPT appsClaude connectors1.00
    28. SOURCE: APPDISCOVERABILITY.COM0.98
    29. MANUFACT.COM1.00
  • 6:02 #12 done7 line(s)

    shot 12·sharpness 1447.9

    1. Manufact1.00
    2. PART 10.92
    3. PART 10.96
    4. WhatMCP Apps0.98
    5. actually are0.96
    6. How an App is declared, rendered, and wired to the model.1.00
    7. MANUFACT.COM1.00
  • 6:33 #13 done18 line(s)

    shot 13·sharpness 1532.0

    1. Manufact1.00
    2. ARCHITECTURE1.00
    3. How an MCP App works1.00
    4. Model1.00
    5. conversation1.00
    6. Host1.00
    7. resources/read1.00
    8. tools/call.0.97
    9. MCP server0.98
    10. reasons, calls tools0.99
    11. ChatGPT·Claude·Cursor1.00
    12. tools + ui:// resources0.97
    13. ui/* over postMessage0.98
    14. Widget1.00
    15. sandboxed iframe1.00
    16. The server exposes the Ul as a ui: //resource. The host renders it in a sandboxed iframe and0.98
    17. relays tool input, results, and the widget's calls back over JSON-RPC.1.00
    18. MANUFACT.COM1.00
  • 7:02 #14 done17 line(s)

    shot 14·sharpness 1917.8

    1. Manufact1.00
    2. ARCHITECTURE1.00
    3. The interaction, end to end1.00
    4. Model1.00
    5. MCP server1.00
    6. Host1.00
    7. Widget1.00
    8. calls a tool1.00
    9. data + a ui:// resource1.00
    10. render the resource in a sandboxed iframe0.99
    11. deliver the tool result1.00
    12. call tools, send messages, update context1.00
    13. Model1.00
    14. MCP server0.96
    15. Host1.00
    16. Widge1.00
    17. MANUFACT.COM1.00
  • 8:03 #15 done27 line(s)

    shot 15·sharpness 1962.6

    1. Manufact1.00
    2. WHAT THEY ARE0.99
    3. Rendered by the host, inline1.00
    4. Same MCP server, with a UI layer on top.0.99
    5. Al assistant0.96
    6. DEMO1.00
    7. The host renders the result inline.1.00
    8. Show me the running shoes you have under $1501.00
    9. Flip Without UI / With UI, and click the cards.0.97
    10. called tool search_catalogue0.99
    11. Without UI1.00
    12. With U0.75
    13. RENDERED BY MCP SERVER0.98
    14. Trailblazer Pro1.00
    15. $1291.00
    16. Cloudstride 21.00
    17. $1451.00
    18. Tempo Racer1.00
    19. $1100.99
    20. Lightweight trail runner1.00
    21. Daily cushioned road shoe1.00
    22. Race-day speed, minimal0.99
    23. weight1.00
    24. Add to cart1.00
    25. Add to cart0.98
    26. Add to cart0.94
    27. MANUFACT.COM1.00
  • 8:13 #16 done6 line(s)

    shot 16·sharpness 1367.5

    1. Manufact1.00
    2. PART 21.00
    3. PART 20.97
    4. Interaction primitives1.00
    5. The ui/* methods the widget and host exchange.1.00
    6. MANUFACT.COM1.00
  • 8:30 #17 done31 line(s)

    shot 17·sharpness 1782.6

    1. Manufact1.00
    2. PRIMITIVE 10.99
    3. ui/update-model-context1.00
    4. The App pushes live state into the model0.99
    5. ui/update-model-context0.98
    6. DEMO1.00
    7. context, with no user message. Add to the1.00
    8. Show me running shoes under $1500.99
    9. cart, and the model already knows what is0.99
    10. WIDGET0.94
    11. Trailblazer Pro1.00
    12. $1291.00
    13. Add to cart0.97
    14. in it on the next turn.0.98
    15. $1451.00
    16. Cloudstride 20.99
    17. Add to cart1.00
    18. Tempo Racer1.00
    19. $1100.84
    20. Add to cart1.00
    21. const { setState } = useWidget();0.99
    22. // pushes ui/update-model-context for you1.00
    23. setState({ cart: ["Tempo Racer"] });1.00
    24. The model knows the cort holds nothing yet.0.98
    25. MODEL CONTEXT0.99
    26. ui/update-model-context1.00
    27. "cart": []0.98
    28. context, so when you ask "what's my fotal?" the model already knows.0.99
    29. Add or remove ifems. The widget pushes the cart to the model with ui/update-model-0.98
    30. Reply...0.96
    31. MANUFACT.COM1.00
  • 9:04 #18 skipped

    shot 18·duplicate of #17

  • 9:10 #19 skipped

    shot 19·duplicate of #17

  • 9:54 #20 skipped

    shot 20·duplicate of #17

  • 10:22 #21 done26 line(s)

    shot 21·sharpness 1485.2

    1. Manufact1.00
    2. PRIMITIVE 21.00
    3. ui/message1.00
    4. The App talks back into the conversation,1.00
    5. ui/message0.98
    6. DEMO1.00
    7. unprompted. A click in the Ul posts a0.99
    8. Show me running shoes under $1501.00
    9. message as if the user sent it.0.99
    10. WIDGET1.00
    11. Trailblazer Pro1.00
    12. $129 - Lightweight trail runner0.97
    13. Learn more0.99
    14. $145 · Daily cushioned rood shoe0.94
    15. Cloudstride 21.00
    16. Learn more1.00
    17. const { sendFollowUpMessage } = useWidget();0.98
    18. sendFollowUpMessage("Add the Tempo Racer");1.00
    19. $110 ·Race-day speed, minimal weight0.96
    20. Tempo Racer1.00
    21. Learn more0.95
    22. CONVERSATION1.00
    23. Click Learn more. The widget colls ui/message and the message appears here,0.99
    24. as if you typed it.0.99
    25. Reply_0.92
    26. MANUFACT.COM1.00
  • 10:44 #22 done25 line(s)

    shot 22·sharpness 1541.7

    1. Manufact1.00
    2. PRIMITIVE 21.00
    3. ui/message1.00
    4. The App talks back into the conversation,1.00
    5. ui/message0.98
    6. DEMO1.00
    7. unprompted. A click in the Ul posts a0.99
    8. Show me running shoes under $1501.00
    9. message as if the user sent it.0.99
    10. WIDGET1.00
    11. Trailblazer Pro1.00
    12. $129-Lightweight trail runner0.99
    13. Learn more1.00
    14. Cloudstride 21.00
    15. $145 · Daily ushioned road shoe0.94
    16. const { sendFollowUpMessage } = useWidget();0.99
    17. sendFollowUpMessage("Add the Tempo Racer");1.00
    18. $110 ·Race-day speed, minimal weight0.98
    19. Tempo Racer1.00
    20. Learn more0.96
    21. CONVERSATION1.00
    22. Tell me more about the Trailblazer Pro1.00
    23. σ Clear0.88
    24. Reply..0.91
    25. MANUFACT.COM1.00
  • 11:03 #23 skipped

    shot 23·duplicate of #22

Transcript

269 cues· 4,364 words· 22,947 chars

  1. 0:01 Hello, y'all.
  2. 0:02 My name is Pietro.
  3. 0:03 I'm the co-founder of Manifect.
  4. 0:05 And today I want to talk to you about MCP apps.
  5. 0:09 Specifically, we're going to talk about the primitives, so how these apps are built, how they work and what they allow you, how to distribute MCP apps, so what is behind the discovery mechanisms of MCP servers and apps in general, and why I think you should care about this because this is how all software will be used.
  6. 0:31 I'm sure most people here listening to this talk know what an MCP is.
  7. 0:36 MCP has been around for quite some time, since 2024.
  8. 0:41 For a full year, it was a very, say, frequent talk amongst developers and companies that were rushing to build these MCP servers.
  9. 0:51 MCP apps are a less familiar concept.
  10. 0:53 They've been around also for quite some time, more or less,
  11. 0:59 since the end of 2025.
  12. 1:03 But when I talk to companies and people in general, I see that many people don't understand what they are and don't understand they can build them.
  13. 1:10 And specifically, they don't know how to distribute them
  14. 1:15 MCP apps, but also they don't know the new way to distribute MCP servers as well.
  15. 1:21 So I hope by the end of this talk, you're going to know about this and you're going to be ready to build your first or iterate on your MCP server and app, and you're going to be able to share it with the world in a more efficient way.
  16. 1:32 It brings you more customers and more users.
  17. 1:38 A little words about me and Manifact, my company.
  18. 1:42 We build open source SDKs and have tools for MCP and MCP Cloud, the Manifact.
  19. 1:50 Open source SDKs, by the name of MCQs, allow developers to build servers
  20. 1:58 and agents in an easier way.
  21. 2:00 So we provide an obstruction over the official SDKs that allows developers to ship faster without worrying about how the spec works beneath.
  22. 2:08 We have 8 million plus downloads across our SDKs and we have 10K stars on GitHub.
  23. 2:15 Also open source, the separate product is the inspector.
  24. 2:20 It's an open source inspector, again, something comparable to the official inspector from the Model Context Protocol maintainers that allows you to test these MCP servers and apps specifically on your local machine.
  25. 2:34 Once you build, once you test it, we have built the cloud for you to ship.
  26. 2:39 So Manifold Cloud is a cloud vertical for MCP.
  27. 2:41 We provide all the primitives for you to be able to ship MCP servers from your GitHub repo and test them immediately, share them with your team, run evals, run publishing checks to make sure that your app is ready to be submitted, and many other features that are completely specific to MCP.
  28. 3:00 Of course, we really believe in MCP.
  29. 3:03 So let's look a bit at the history of MCP and how did we get there.
  30. 3:07 So MCP was launched in 2024.
  31. 3:09 And by the end of May 2025, Ido Solomon and Lied Josef started working on MCP UI, which is this way to kind of MCP servers return UI components.
  32. 3:21 Then this was like many people started talking about MCP UI because this is like such a great opportunity to ship UI with your MCP servers to agents.
  33. 3:32 be these interactive experiences where the agent is calling tools, but is showing UIs to the user.
  34. 3:37 This created a lot of movement.
  35. 3:39 Many people enjoyed this proposal.
  36. 3:41 And the chat GPT at some point released the app SDK, which is a way to create these interactive MCP apps, basically MCP servers that also return UI elements.
  37. 3:55 Also, at the end of 2025, quite silently,
  38. 4:02 and their stores for MCP.
  39. 4:05 These stores allow you to submit your MCP server and have a one-click install experience for your users.
  40. 4:12 For the most time, these stores were closed and it says they were designed and only allowed for design partners, but things have changed and I'm happy
  41. 4:29 In January 2026, MCPUI converted to MCP apps.
  42. 4:37 MCP apps is now the official extension of the model context protocol that allows to return UI elements within MCP servers.
  43. 4:46 So this timeline is kind of an explanation of how the protocol evolved.
  44. 4:50 And I think two major things happened in this timeline.
  45. 4:54 First, MCP apps.
  46. 4:56 MCP servers are not only returning JSON, and that allows much richer experiences.
  47. 5:02 And the second thing, maybe even bigger, is that the stores opened.
  48. 5:06 This was a great, huge move by the model providers, Anthropic, OpenAI, Bursar, basically every LLM client out there, to say, oh, MCP is the way, and we want to have a way for people to publish vetted and quality MCP servers
  49. 5:27 experience.
  50. 5:31 This is the situation right now with the stores.

Open at this second