vidtheque.

read-only demo

Videos v3Fr2JR47KA

The Future of MCP — David Soria Parra, Anthropic

index_state ready data_status ok

AI Engineer· published 2026-04-19· 0:18:45· en-US· indexed 2026-08-10 19:45

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:05, 1 of 1 keyframes kept
  2. Shot 1, 0:05 to 0:09, 1 of 1 keyframes kept
  3. Shot 2, 0:09 to 0:14, 1 of 1 keyframes kept
  4. Shot 3, 0:14 to 0:16, 1 of 1 keyframes kept
  5. Shot 4, 0:16 to 0:17, 1 of 1 keyframes kept
  6. Shot 5, 0:17 to 0:47, 1 of 1 keyframes kept
  7. Shot 6, 0:47 to 0:55, 1 of 1 keyframes kept
  8. Shot 7, 0:55 to 1:35, 1 of 1 keyframes kept
  9. Shot 8, 1:35 to 1:41, 1 of 1 keyframes kept
  10. Shot 9, 1:41 to 1:53, 1 of 1 keyframes kept
  11. Shot 10, 1:53 to 1:58, 1 of 1 keyframes kept
  12. Shot 11, 1:58 to 2:26, 1 of 1 keyframes kept
  13. Shot 12, 2:26 to 2:28, 1 of 1 keyframes kept
  14. Shot 13, 2:28 to 2:54, 1 of 1 keyframes kept
  15. Shot 14, 2:54 to 3:12, 1 of 1 keyframes kept
  16. Shot 15, 3:12 to 3:45, 1 of 1 keyframes kept
  17. Shot 16, 3:45 to 3:58, 1 of 1 keyframes kept
  18. Shot 17, 3:58 to 4:00, 1 of 1 keyframes kept
  19. Shot 18, 4:00 to 4:49, 1 of 1 keyframes kept
  20. Shot 19, 4:49 to 5:18, 1 of 1 keyframes kept
  21. Shot 20, 5:18 to 5:45, 1 of 1 keyframes kept
  22. Shot 21, 5:45 to 6:02, 1 of 1 keyframes kept
  23. Shot 22, 6:02 to 6:34, 1 of 1 keyframes kept
  24. Shot 23, 6:34 to 7:06, 1 of 1 keyframes kept
  25. Shot 24, 7:06 to 7:13, 1 of 1 keyframes kept
  26. Shot 25, 7:13 to 7:26, 1 of 1 keyframes kept
  27. Shot 26, 7:26 to 8:02, 1 of 1 keyframes kept
  28. Shot 27, 8:02 to 8:34, 1 of 1 keyframes kept
  29. Shot 28, 8:34 to 9:07, 0 of 1 keyframes kept
  30. Shot 29, 9:07 to 9:39, 1 of 1 keyframes kept
  31. Shot 30, 9:39 to 9:43, 1 of 1 keyframes kept
  32. Shot 31, 9:43 to 9:51, 1 of 1 keyframes kept
  33. Shot 32, 9:51 to 10:05, 1 of 1 keyframes kept
  34. Shot 33, 10:05 to 10:29, 1 of 1 keyframes kept
  35. Shot 34, 10:29 to 10:32, 1 of 1 keyframes kept
  36. Shot 35, 10:32 to 10:59, 1 of 1 keyframes kept
  37. Shot 36, 10:59 to 11:26, 0 of 1 keyframes kept
  38. Shot 37, 11:26 to 11:59, 1 of 1 keyframes kept
  39. Shot 38, 11:59 to 12:08, 1 of 1 keyframes kept
  40. Shot 39, 12:08 to 12:13, 1 of 1 keyframes kept
  41. Shot 40, 12:13 to 12:39, 1 of 1 keyframes kept
  42. Shot 41, 12:39 to 13:09, 1 of 1 keyframes kept
  43. Shot 42, 13:09 to 13:38, 1 of 1 keyframes kept
  44. Shot 43, 13:38 to 13:41, 1 of 1 keyframes kept
  45. Shot 44, 13:41 to 13:51, 1 of 1 keyframes kept
  46. Shot 45, 13:51 to 14:00, 1 of 1 keyframes kept
  47. Shot 46, 14:00 to 14:24, 1 of 1 keyframes kept
  48. Shot 47, 14:24 to 15:03, 1 of 1 keyframes kept
  49. Shot 48, 15:03 to 15:17, 1 of 1 keyframes kept
  50. Shot 49, 15:17 to 15:52, 1 of 1 keyframes kept
  51. Shot 50, 15:52 to 16:27, 1 of 1 keyframes kept
  52. Shot 51, 16:27 to 16:34, 1 of 1 keyframes kept
  53. Shot 52, 16:34 to 16:59, 1 of 1 keyframes kept
  54. Shot 53, 16:59 to 17:24, 1 of 1 keyframes kept
  55. Shot 54, 17:24 to 17:49, 1 of 1 keyframes kept
  56. Shot 55, 17:49 to 18:05, 1 of 1 keyframes kept
  57. Shot 56, 18:05 to 18:13, 1 of 1 keyframes kept
  58. Shot 57, 18:13 to 18:28, 1 of 1 keyframes kept
  59. Shot 58, 18:28 to 18:44, 1 of 1 keyframes kept
  60. Shot 59, 18:44 to 18:45, 0 of 1 keyframes kept

60 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
176
whisperx 176
chunks
33
from 176 cues
keyframes
57
kept of 60 captured
frames with text
57
1,479 lines read
chapters
12
from the source metadata
keyframe bytes
5.7 MB
word timings on 176 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 01:05 1m 56s
stt done — 2026-08-10 01:07 21s
chunk done — 2026-08-10 01:07 0s
text_embed done — 2026-08-10 19:45 1s
keyframe done — 2026-08-10 01:07 1m 33s
ocr done — 2026-08-10 01:09 28s
frame_embed done — 2026-08-10 19:45 9s

Frames, and what the machine read

  • 0:03 #0 done2 line(s)

    shot 0·sharpness 666.1

    1. AlEngineer0.98
    2. EUROPE1.00

  • 0:07 #1 done2 line(s)

    shot 1·sharpness 827.5

    1. PRESENTING SPONSOR0.99
    2. Google DeepMind1.00

  • 0:13 #2 done3 line(s)

    shot 2·sharpness 910.4

    1. PLATINUM SPONSORS0.98
    2. # Braintrust0.96
    3. WorkOS OpenAI0.95

  • 0:16 #3 done5 line(s)

    shot 3·sharpness 594.0

    1. CO-CREATOOF MCPMEMBER OF1.00
    2. ANTHROP1.00
    3. neer1.00
    4. D BY0.96
    5. epMind1.00

  • 0:17 #4 done14 line(s)

    shot 4·sharpness 819.4

    1. ntrust0.92
    2. /Engineer0.94
    3. BUMOPE0.96
    4. Microsoft1.00
    5. The Future of MCP0.98
    6. er.dev1.00
    7. E.gineer0.84
    8. Al Engineer·London·April 20260.98
    9. EUROPE1.00
    10. neer1.00
    11. tailscale1.00
    12. David Soria Parra · Member of Technical Staff0.99
    13. AlEngineer0.98
    14. EUROPE1.00

  • 0:32 #5 done33 line(s)

    shot 5·sharpness 1495.5

    1. Claude File Edit View Developer Window Help0.99
    2. QThu Feb5 07:260.86
    3. FEBRUARY0.98
    4. ●□0.56
    5. Chat1.00
    6. Cowork1.00
    7. Code1.00
    8. Cute Raspberry Pi 5 drawing in Excalidraw0.96
    9. raintrus1.00
    10. AIEngineea0.90
    11. Excalidraw create_view0.98
    12. 小0.99
    13. London1.00
    14. Expect ligt rain in0.99
    15. the next hour0.93
    16. ll…0.60
    17. Microsc0.94
    18. gger.d0.93
    19. AEngineer1.00
    20. Ethernet1.00
    21. tailsca1.00
    22. RJ451.00
    23. Reply....0.94
    24. 十0.91
    25. Opus 4.60.93
    26. Claude is Al and can make mistakes. Please double-check responses.0.99
    27. AlEngineer0.98
    28. THE FUTURE OF MCP0.98
    29. EUROPE1.00
    30. PRESENTED BY1.00
    31. Google DeepMind1.00
    32. DAVID SORIA PARRA / Member of Technical Staff1.00
    33. ANTHROP\C1.00

  • 0:54 #6 done3 line(s)

    shot 6·sharpness 719.9

    1. What you just saw0.97
    2. A agent shipping its own interface —1.00
    3. through a protocol.1.00

  • 1:23 #7 done21 line(s)

    shot 7·sharpness 1701.3

    1. gineer1.00
    2. together0.98
    3. OPE1.00
    4. What you just saw0.97
    5. enA0.93
    6. ""Engineer0.96
    7. UROPE1.00
    8. A agent shipping its own interface —0.99
    9. through a protocol.0.99
    10. DENTR1.00
    11. OPL1.00
    12. Tes0.99
    13. ngineer1.00
    14. UROPE1.00
    15. AlEngineer0.96
    16. THE FUTURE OF MCP0.99
    17. EUROPE1.00
    18. PRESENTED BY1.00
    19. Google DeepMind1.00
    20. DAVID SORIA PARRA / Member of Technical Staff0.99
    21. ANTHROP\C1.00

  • 1:40 #8 done14 line(s)

    shot 8·sharpness 1883.7

    1. AlEngineer0.95
    2. EUROPE1.00
    3. How we got here0.97
    4. A year ago this was a spec doc,1.00
    5. local only, with little more than just tools.1.00
    6. AB0.66
    7. Mia0.97
    8. AlEngineer0.98
    9. THE FUTURE OF MCP0.99
    10. EUROPE1.00
    11. PRESENTED BY1.00
    12. Google DeepMind1.00
    13. DAVID SORIA PARRA / Member of Technical Staff0.99
    14. ANTHROP\C1.00

  • 1:42 #9 done18 line(s)

    shot 9·sharpness 2096.9

    1. Zed0.98
    2. AlEnginee0.95
    3. NEng0.90
    4. WorkOs0.94
    5. How we got here1.00
    6. SAFE1.00
    7. rize1.00
    8. NIEn0.86
    9. A year ago this was a spec doc,0.99
    10. :neouy0.70
    11. local only, with little more than just tools.0.99
    12. AlEngineer0.98
    13. THE FUTURE OF MCP1.00
    14. EUROPE1.00
    15. PRESENTED BY1.00
    16. Google DeepMind1.00
    17. DAVID SORIA PARRA / Member of Technical Staff0.98
    18. ANTHROP\C1.00

  • 1:54 #10 done24 line(s)

    shot 10·sharpness 2036.0

    1. AEngineer0.96
    2. EUROPE1.00
    3. How we got here0.97
    4. Braintrus1.00
    5. AlEnginee0.95
    6. EUROPE1.00
    7. AIEngineer0.96
    8. Micros1.00
    9. A year ago this was a spec doc,0.97
    10. EUROPE1.00
    11. local only, with little more than just tools.1.00
    12. rigger.de1.00
    13. AlEnginee0.98
    14. EUROPE1.00
    15. AIEngineer0.97
    16. tailsc1.00
    17. EUROPE1.00
    18. AlEngineer0.97
    19. THE FUTURE OF MCP0.99
    20. EUROPE1.00
    21. PRESENTED BY1.00
    22. Google DeepMind1.00
    23. DAVID SORIA PARRA / Member of Technical Staff0.99
    24. ANTHROP\C1.00

  • 2:12 #11 done35 line(s)

    shot 11·sharpness 1773.7

    1. Engineer1.00
    2. EUROPE1.00
    3. Twelve months of MCP0.97
    4. raintrus1.00
    5. Engineer1.00
    6. OPE0.85
    7. Engineer1.00
    8. Microso0.95
    9. EUROPE1.00
    10. igger.d0.98
    11. IEngineer0.95
    12. Nov '240.97
    13. Mar'251.00
    14. Jun '250.97
    15. Sep '250.98
    16. Dec '250.93
    17. Q1 '260.95
    18. EUROPE1.00
    19. Open-sourced1.00
    20. Remote servers1.00
    21. Authorization1.00
    22. Elicitation1.00
    23. Tasks1.00
    24. MCP Apps0.96
    25. Engineer1.00
    26. tailsca1.00
    27. EUROPE1.00
    28. Hundreds of clients ·Thousands of servers·one spec0.99
    29. AlEngineer0.98
    30. THE FUTURE OF MCP0.97
    31. EUROPE1.00
    32. PRESENTED BY1.00
    33. Google DeepMind1.00
    34. DAVID SORIA PARRA / Member of Technical Staff1.00
    35. ANTHROP\C1.00

  • 2:27 #12 done21 line(s)

    shot 12·sharpness 1556.8

    1. Twelve months of MCP0.98
    2. Nov '240.95
    3. Mar'251.00
    4. Jun '250.98
    5. Sep '250.93
    6. Dec '250.93
    7. Q1 '260.92
    8. Open-sourced1.00
    9. Remote servers0.98
    10. Authorization1.00
    11. Elicitation1.00
    12. Tasks1.00
    13. MCP Apps1.00
    14. Hundreds of clients ·Thousands of servers·one spec0.99
    15. AlEngineer0.98
    16. THE FUTURE OF MCP0.98
    17. EUROPE1.00
    18. PRESENTED BY1.00
    19. Google DeepMind1.00
    20. DAVID SORIA PARRA / Member of Technical Staff0.99
    21. ANTHROP\C1.00

  • 2:33 #13 done16 line(s)

    shot 13·sharpness 1195.2

    1. Braintrust1.00
    2. AlEnginee0.98
    3. NEngineer0.96
    4. EUROPE0.98
    5. 110m+1.00
    6. Snork1.00
    7. SDK downloads per month. Python, Typescript, and the rest1.00
    8. NEngi0.80
    9. EURO.0.94
    10. AlEngineer0.97
    11. THE FUTURE OF MCP0.97
    12. EUROPE1.00
    13. PRESENTED BY1.00
    14. Google DeepMind1.00
    15. DAVID SORIA PARRA / Member of Technical Staff1.00
    16. ANTHROP\C1.00

  • 3:10 #14 done16 line(s)

    shot 14·sharpness 1344.5

    1. 圆Zed0.86
    2. GoogleDeepMi0.98
    3. herai0.87
    4. NB0.50
    5. ATrigs0.94
    6. 110m+1.00
    7. ize0.91
    8. AIB0.71
    9. SDK downloads per month. Python, Typescript, and the rest1.00
    10. AlEngineer0.98
    11. THE FUTURE OF MCP0.98
    12. EUROPE1.00
    13. PRESENTED BY1.00
    14. Google DeepMind1.00
    15. DAVID SORIA PARRA / Member of Technical Staff0.98
    16. ANTHROP\C1.00

  • 3:16 #15 done20 line(s)

    shot 15·sharpness 1204.9

    1. ineer0.98
    2. ether.ai0.97
    3. ineer0.99
    4. o4j0.83
    5. 110m+1.00
    6. ineer1.00
    7. rize1.00
    8. igineer0.98
    9. JROPE0.93
    10. SDK downloads per month. Python, Typescript, and the rest1.00
    11. ineer1.00
    12. ailscale1.00
    13. PE0.99
    14. AlEngineer0.98
    15. THE FUTURE OF MCP0.97
    16. EUROPE1.00
    17. PRESENTED BY1.00
    18. Google DeepMind1.00
    19. DAVID SORIA PARRA / Member of Technical Staff1.00
    20. ANTHROP\C1.00

  • 3:54 #16 done24 line(s)

    shot 16·sharpness 1534.2

    1. IEngineer0.96
    2. EUROPE1.00
    3. But this is just the start0.99
    4. together1.00
    5. AlEngineer0.95
    6. 一EUROPE一0.89
    7. 2026 is the year agents go to production.1.00
    8. Engineer1.00
    9. neo40.92
    10. EUROPE1.00
    11. arize0.95
    12. AIEngineer0.92
    13. EUROPE1.00
    14. Engineer1.00
    15. tailsci0.98
    16. EUROPE1.00
    17. AlEngineer0.98
    18. THE FUTURE OF MCP0.97
    19. EUROPE1.00
    20. PRESENTED BY1.00
    21. Google DeepMind1.00
    22. #Braintrust0.97
    23. WorkOS1.00
    24. OpenAl0.92

  • 3:58 #17 done32 line(s)

    shot 17·sharpness 1479.3

    1. NEngineer0.95
    2. EUROPE1.00
    3. The curve1.00
    4. We are (still) early1.00
    5. together0.92
    6. AlEnginee0.95
    7. EUROPE1.00
    8. AEnginee0.94
    9. neo40.93
    10. EUROPE1.00
    11. arize1.00
    12. AlEnginee0.95
    13. we are here1.00
    14. EUROPE0.99
    15. early 20261.00
    16. AEngineer0.98
    17. tailsc1.00
    18. EUROPE1.00
    19. Demos1.00
    20. Coding agents1.00
    21. Knowledge work1.00
    22. 20241.00
    23. 20251.00
    24. 20261.00
    25. AlEngineer0.95
    26. THE FUTURE OF MCP0.99
    27. EUROPE1.00
    28. PRESENTED BY1.00
    29. Google DeepMind1.00
    30. # Braintrust0.94
    31. WorkOS1.00
    32. OpenAl0.94

  • 4:20 #18 done10 line(s)

    shot 18·sharpness 591.5

    1. The curve1.00
    2. We are (still) early1.00
    3. we are here0.99
    4. early 20261.00
    5. Demos1.00
    6. Coding agents0.97
    7. Knowledge work1.00
    8. 20241.00
    9. 20251.00
    10. 20261.00

  • 5:09 #19 done26 line(s)

    shot 19·sharpness 1099.7

    1. AlEngineer0.99
    2. EUROPE1.00
    3. The curve0.98
    4. We are (still) early1.00
    5. IEngineer0.95
    6. Wor'C0.80
    7. EUROPE1.00
    8. le DeepN0.89
    9. AIEngineer0.97
    10. EUROPE1.00
    11. ariz1.00
    12. IEngineer0.96
    13. we are here1.00
    14. EUROPE1.00
    15. early 20261.00
    16. eNCORD0.98
    17. AlEngineer0.98
    18. EUROPE1.00
    19. Demos1.00
    20. Coding agents0.99
    21. Knowledge work1.00
    22. 20240.99
    23. 20251.00
    24. 20261.00
    25. Engineering the future of Al0.98
    26. AlEngineer0.98

  • 5:26 #20 done34 line(s)

    shot 20·sharpness 1749.0

    1. EUROPE1.00
    2. The connectivity stack0.98
    3. Using the right tool1.00
    4. gineer0.98
    5. Braintrust1.00
    6. enAl0.98
    7. AIEngineer0.97
    8. EUROPE0.99
    9. M1.00
    10. <1>0.99
    11. gineer1.00
    12. Microsoft1.00
    13. OPE0.99
    14. Skills1.00
    15. MCP1.00
    16. CLI / Computer0.96
    17. use1.00
    18. Modal1.00
    19. AlEngineer0.97
    20. Domain knowledge captured as reusable1.00
    21. The integration protocol. Semantics, governance,0.98
    22. General access to existing systems. Discoverable,0.99
    23. EUROPE1.00
    24. instructions.1.00
    25. and reach across boundaries.1.00
    26. composable, Unix-style.1.00
    27. These compose. Agents in 2026 use all of them.0.99
    28. AlEngineer0.97
    29. THE FUTURE OF MCP0.99
    30. EUROPE1.00
    31. PRESENTED BY1.00
    32. Google DeepMind1.00
    33. DAVID SORIA PARRA / Member of Technical Staff1.00
    34. ANTHROP\C1.00

  • 6:00 #21 done32 line(s)

    shot 21·sharpness 2013.0

    1. AlEngine0.95
    2. EUROPE1.00
    3. Command Line1.00
    4. Unix Philosophy for Agents1.00
    5. AlEngineer0.98
    6. Work1.00
    7. EUROPE0.99
    8. Composition1.00
    9. Progressive Discovery1.00
    10. gle Dee,0.90
    11. AlEngine0.95
    12. EUROPE1.00
    13. Bash Tool composition0.99
    14. Discover usage when needed0.98
    15. Pipes, redirects, scripting1.00
    16. Models can explore and discover CLls when needed0.98
    17. AIEngineer0.95
    18. ariz1.00
    19. EUROPE1.00
    20. Learned1.00
    21. eNCORD0.94
    22. AlEngine0.96
    23. EUROPE1.00
    24. Pretraining1.00
    25. Particularly strong when pre-trained, e.g. gh, git, etc.0.99
    26. AlEngineer0.97
    27. THE FUTURE OF MCP0.99
    28. EUROPE1.00
    29. PRESENTED BY1.00
    30. Google DeepMind1.00
    31. DAVID SORIA PARRA / Member of Technical Staff0.99
    32. ANTHROP\C1.00

  • 6:27 #22 done21 line(s)

    shot 22·sharpness 1965.0

    1. pogle Deoind0.85
    2. AlEng0.90
    3. Command Line1.00
    4. Unix Philosophy for Agents1.00
    5. ATrigg0.99
    6. Composition1.00
    7. Progressive Discovery1.00
    8. Bash Tool composition1.00
    9. Discover usage when needed0.98
    10. Pipes, redirects, scripting1.00
    11. Models can explore and discover CLls when needed0.98
    12. Learned1.00
    13. Pretraining1.00
    14. Particularly strong when pre-trained, e.g. gh, git, etc.1.00
    15. AlEngineer0.97
    16. THE FUTURE OF MCP0.99
    17. EUROPE1.00
    18. PRESENTED BY1.00
    19. Google DeepMind1.00
    20. DAVID SORIA PARRA / Member of Technical Staff0.99
    21. ANTHROP\C1.00

  • 7:03 #23 done31 line(s)

    shot 23·sharpness 2338.5

    1. Google1.00
    2. AlEngineer0.99
    3. EUROPE1.00
    4. MCP1.00
    5. Connective Tissue1.00
    6. toge"0.84
    7. AIEn0.89
    8. Semantics1.00
    9. Decoupling1.00
    10. Tools, Resources, Tasks & Co0.99
    11. Platform independence1.00
    12. Typed capabilities and context, discoverable on demand.0.99
    13. Don't rely on a given platform execution environment/1.00
    14. sandbox.0.95
    15. AlEng0.90
    16. EURO1.00
    17. Enterprise1.00
    18. Experiments1.00
    19. Auth, Governance & Policies1.00
    20. MCP Apps, Skills over MCP (soon)1.00
    21. Λan0.75
    22. Things, big companies care about. Filtering, auditing,1.00
    23. Ship interactive Ul or skills through the wire.0.98
    24. observability1.00
    25. AlEngineer0.98
    26. THE FUTURE OF MCP0.99
    27. EUROPE1.00
    28. PRESENTED BY1.00
    29. Google DeepMind1.00
    30. DAVID SORIA PARRA / Member of Technical Staff0.99
    31. ANTHROP\C1.00

Transcript

176 cues· 3,149 words· 17,169 chars

  1. 0:15 Well, welcome.
  2. 0:18 Let's get started.
  3. 0:21 This is an MCP application.
  4. 0:25 That's an agent shipping its own interface, not through a plug-in, not through an SDK, not rendered on the fly by the model on the client side, or hard-coded into the product.
  5. 0:37 That is something that is served over an MCP server.
  6. 0:40 And you can take the server, put it into Cloud, you can put it into ChatGPT, you can put it into VS Code cursor, and it will just fucking work.
  7. 0:50 And that, I think, is kind of cool.
  8. 0:53 Because for doing that, you need something that a lot of things that we want in the ecosystem do not offer.
  9. 0:59 You need semantics.
  10. 1:00 You need to have both sides, the client and the server, to understand what each side is talking, to understand how you render this, understand that there is a UI coming.
  11. 1:10 And for that, you need a protocol.
  12. 1:13 And the best part about this?
  13. 1:15 An MCP server doesn't just ship an app or can ship an app.
  14. 1:19 It can also ship tools with it.
  15. 1:21 And so you can interact with it with the application as a human.
  16. 1:25 And you can have the model interact with it through tools, which is, I think, a very unique thing that I think we have not explored much just yet.
  17. 1:34 OK, but let's quickly rewind a little bit from what I think is a really cool glimpse into the future of MCP into over a year ago, 18 months, an eternity in AI lifecycle.
  18. 1:48 All of this did not exist.
  19. 1:50 There was just a little spec document, a few SDKs, mostly written by Claude, local only, with little more than just tools.
  20. 1:59 And in that last 18 or 12 months, you guys have been absolutely crazy building stuff, building servers, building a crazy ecosystem around this.
  21. 2:08 And we on our side have been busy
  22. 2:11 taken this local only thing, added remote capabilities, added centralized authorization, added new primitives like elicitation and tasks, and last but not least, added new experimental features to the protocol like the MCP applications that you've just seen.
  23. 2:30 And in the meantime, we have reached, I think, a really cool milestone because, again, all of you have been absolutely crazy building, building, and building, of course, luckily with the help of a bunch of agents.
  24. 2:42 We're now at 110 million.
  25. 2:45 monthly downloads.
  26. 2:46 And that's just, of course, not us using it in our clients and servers.
  27. 2:50 That's like OpenAI's agents SDK, that's Google's ADK, that's Langchain, thousands of frameworks and tools that you might have never ever heard of it, pulling it in as a dependency, which means there's one common standard that all of us have at our disposal to speak to each other.
  28. 3:09 Just a bit for context, React, one of the most successful open source projects probably of the last decades, took roughly double the amount of time to reach that download volume.
  29. 3:20 And in the meantime, of course, you all have been building really, really cool servers from little toy projects of WhatsApp servers and Blender servers to building SaaS integrations like Linear, Slack, and Notion that are really powering what everyone does every day when they use MCPs.
  30. 3:34 But most importantly, the vast majority of MTP server most of all of us have built are behind closed doors, connecting company systems to agents and AI applications.
  31. 3:46 But I still think this is just the absolute beginning of where we are.
  32. 3:51 Because I think 2025 was all about exploring, and 2026 is all about putting these agents into production.
  33. 3:59 Because if you really think about, in my mind, 2024, we just built a bunch of demos and showed cool stuff to people, and there was a little bit of a buzz there.
  34. 4:09 2025 was really all about coding agents.
  35. 4:10 But coding agents, if you really think about it, are the most
  36. 4:14 ideal scenario for an agent.
  37. 4:16 It's local, it's verifiable, you can call a compiler, like you have a developer who can fix shit if it goes wrong in front of the computer, and you can display a TUI interface and the user's quite happy.
  38. 4:30 But I think now, with the capabilities of the model increasing, we are going into a new era, which I think this year we will see this start, where we're not just doing coding agents.
  39. 4:40 We're going to have general agents that will do real knowledge worker stuff, like things a financial analyst want to do, a marketing person want to do.
  40. 4:50 And they need one thing in particular.
  41. 4:54 They don't need a local agent that calls a compiler.
  42. 4:56 What they need is something that could connect to like five SAS applications and a shared drive, because the most important part for them, for an agent, is connectivity.
  43. 5:06 And in my mind, connectivity is not one thing.
  44. 5:09 If someone tells you there's one solution to all your connectivity problems, be it computer use, be it CLI's, be it MCP, they are probably pretty wrong.
  45. 5:17 Because the right thing, of course, is that it always means it depends.
  46. 5:22 And there's a real big connectivity stack.
  47. 5:25 And there's the right tool for the right job.
  48. 5:28 And in my mind, there are three major things that you want to consider building an agent in 2026.
  49. 5:32 It's skills, MCP, and of course, like CLI or computer use, depending on your use case.
  50. 5:38 And they have three very distinct things that they can do and three different things you want to consider when you build your agent.

Chapters

  1. 0:00 Introduction and the vision for MCP applications
  2. 1:34 Looking back at the evolution of the MCP ecosystem over the last 18 months
  3. 2:30 Ecosystem growth and adoption milestones
  4. 3:46 Moving from exploration in 2025 to production in 2026
  5. 5:07 The 2026 connectivity stack: Skills, MCP, and CLI/Computer use
  6. 7:47 Improving client harnesses: Progressive Discovery
  7. 9:39 Programmatic tool calling and agent orchestration
  8. 12:00 Best practices for designing agents and server authors
  9. 13:42 Future roadmap for the MCP protocol and core improvements
  10. 15:23 Strategic integrations and enterprise features
  11. 16:32 Upcoming extension mechanisms and skills over MCP
  12. 17:15 Conclusion and call for community feedback

Open at this second