Slides: MCP is all you need — Samuel Colvin, Pydantic
Source Video
MCP is all you need — Samuel Colvin, Pydantic
Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.
Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides
OCR text:
INNOVATIONPARTNER
aws
PLATINUMSPONSORS
Graphite
WWindsurf
MongoDB
daily
augment code
WorkOs
OCR text:
aily Worle'- AlEngineer Graphite Wor
d'sFair Wrld' Eair AlEngineer
World's Fair AIEngineer
ngbase PRESENTED BY Microsoft Wor
OCR text:
AI Engineer
World's Fair
OCR text:
what
MCP is all you need
Inspired by Jason Liu's talkPydantic is all you needand“Pydantic is still all you need
Same idea:that lots ofpeople are over complicating things
AIE
Same unrealistic title-no one is seriously claiming MCP can do everything
What I an saying is:
HCP can do a lot of multi-agent communications-specifically autonomous agents.
World'sFair
rite
Wo
Weights
Microsoft
aws
smol?
World'sFair
OCR text:
非 McP for multi-agent communication
The way most people describe it:
MCP
tool
AIE
Agent
tool
tool
World'sFair
World:
SEI
Microsoft
aws
smol?
World'sFair
OCR text:
# sampling
Give MCP servers the ability to make requests to LLMs via the client.
（Powerfulfeature of McP,but not widely supported)
AIE
LLR
RCPclient
MCPserver
LUNCall
LLN tool call response
toolcall
saapling'createmessage*
LURcall
LLK text response
sanpling response
toolcall response
RCPclient
World'e
daily
Wor
lndd'
cair
Microsoft
aws
smol?
World'sFair
OCR text:
meu MCP client MCP server
LLM call
a EEEEEEEEEEEUUEE EEE EEIEEEEEE EERE URE EERE NUE
LLM tool call response
tool call
sampling "create message"
scar EEE IETE EEE En
LLM call
Sa 4 —— eee
LLM text response
4]
sampling response
tool call response
ey ————— ——t
OCR text:
# sampling
Give MCP servers the ability to make requests to LLMs via the client.
(Powerful feature of MCP,but not widely supported*)
LLM
MCPclient
HCP server
LLM call
LLMtoolcallresponse
toolcall
sanplingcreatemessage
LLM call
LLN text response
osuodsax Bugtdaes
toolcallresponse
LLM
HCPclient
MCP server
OCR text:
G ledPrevee ft fii Selection ew Oo Widow Map Q &@ © Reet © m © A Qe wstan ics
eo
wo a ce oy mands)
EP are (re a
is an ee 2 Sat es a ae ea car
vbr ts gee 5 ae Sey ge A ar mae
Pe
ae ee
2 sid ge agg eer tar eC eee ae ec ee eee
eet ae ee
rey
B a on eer OP ae re one rca a a
f 5 3
3G O&
OCR text:
ited Proview! {te Cai) Seiscuan Wires ie Wires hae, Ce Ot Sm Gs matric
ar
PTs ory
* eee ee ae ee eK re ee ee ee
Cn es ata
Been
aoe a aes a | ae UG eee oem oo ae ki MEL OE
ger * 9 hoe gel) “og a a Pa ee a eee en a ee sre a?
et | Pe i a arr a ar ae
on ee crea cc a
ee ee ee en co
% ; at F
H 5 FI
. _O& uw
OCR text:
ZedPrevien Fle tdt Go
README.md settings
118
120 121 119 122 123 async def run_query(ctx:RunContext[Deps],sql:str)→str: @pypi_agent.output_validator m=re.search（r ifm: remove sql.. \w\n(.?)...,,sql,flags=re.S)
124 sql=m.group（1).strip（)
125
126 logfire.info('running{sql}',sql=sql）
127 128 129 await ctx.deps.mcp_context.log（'info','runningquery') iff'from{table_name}'not in sql.lower(）: raise ModelRetry（f'Query must be against the{table_name?table')
132 130 131 try: rows=query_job.result（) query_job=client.query（sql)
133 except BadRequest as e:
134
135 136 await ctx.deps.mcp_context.log('info','query successful') raise ModelRetry（f'Invalid query:{e}')from e
1L
OCR text:
G ledPrevwe ft fai Selection ew Oo Wridow Map Q @ © Rt © mw F A Qe wesaans sto
oad
es Peg t
ates OCs eee MO LESSEE 0
oa ara whe 4
vA . i ore bre eae OR
ors ee oy ,
end Ca cone 5
. peg et 2 btu. der.ig, ee) oe ees
ert aan Fa Lear 5 7 7 A A , s
al Se a er me ey a
ee ene re 9g co *, 9 8 5
wat Syl oe :
' im ; : a Ed
aC a): oe,
OCR text:
ZedPrevien Fle tdt
README.md setings
class Deps
124 123 125 ifm: sql=m.group（1).strip（）
126 logfire.info('running {sql}',sql=sql）
127 128 await ctx.deps.mcp_context.log（'info','runningquery') iff'from{table_name}'notin sql.lower(）:
129
132 130 131 133 134 135 try: except BadRequest as e: rows=query_job.result（) query_job=client.query(sql) await ctx.deps.mcp_context.log（'warning','query error retrying') raise ModelRetry（f'Invalid query:{e}')from e
136 await ctx.deps.mcp_context.log('info',‘query successful')
137 data=[dict（row) for row in rows] #type:ignore
138 return format_as_xml（data,item_tag='row',include_root_tag=False)
139
140
141 mcp=FastMCP（'PyPI query',log_level='WARNING'）
1L
OCR text:
ZedPreview Fle Edt 四
READMEmd setings
pypi_ncp_setves.py 138 139 137 data=[dict（row）for row inrows]type:ignore return format_as_xml（data,item_tag='row',include_root_tag=False)
140
141 142 mcp =FastMCP（'PyPI query',log_level='WARNING'）
143
144 145 146 async def pypi_downloads(question:str,ctx:Context[ServerSession,None]) → str: @mcp.tool() aboutpackagedownloads."""
147 result=await pypi_agent.run(question,model=MCPSamplingModel（session=ctx.session),
148 deps=Deps(ctx)) return result.output
149
150 152 151 153 if_name #from devtoolsimport debug mcp.run() main
1L
OCR text:
data=[dict（row)for row in rows]# type:ignore
return format_as_xml（data,item_tag='row'.include_root_tag=False)
mcp=FastMCP（'PyPI query',1og_level='WARNING'）
AIE
@mcp.tool()
async def pypi_downloads(question:str,ctx:Context[ServerSession,None])→ str:
""Analyze downloads of packages from the Python package index PyPI to answer questions
about package downloads."*"I
result = await pypi_agent.run(question,model=McPSamplingModel（session=ctx.session),
deps=Deps(ctx))
return result.output
if
name
main
run()
om devtoolsimport debug
World's
vellum
W
Microsoft
aws
Alndun
Cal
smol?
World'sFair
OCR text:
de-add_date()
2Bdefaddetiarhef'Today is{date.today():%Y-%m-%d}
return f'Today is{date.today():%Y-%m-%d3
AIE
async def main():
asyncwith libs_agent.run_mcp_servers():
Samuel Colvin,6 houzs ago
33aSync def result（await libs_agent.run(How many times has pydantic been downloaded this year）
asprintixesultpusputgent.run_mcp_servers(
result= await libs_agent.run('How many times has pydantic been
if
naownlaadedathis:year')
prindorteasyhtiooutput)
asyncio.run（main())
nama
libs agent run
MCP request:tools/list
chat gpt-4o
World's
ThvellumgepydWichas been down1oaded 1,687,008,160
1L
times
this year
Vdone
Microsoft
aws
Cair
smol"
World'sFair
1L
OCR text:
LS coveene: re, shas, irom rooms Demernarees <fonfere: 7am mm ee OSG me tanets
oe me eee
a ee fie : i
a Coe ee ee ee ee ee ee F a oe
“rie cer'] MCP request: initialize
Sree] WRB Lads agent run cangang?s
(8) RP request. initialize
face oc} B20 tide agent tun
iN] tT PCP request: tools/list
MCP soreer handle request: toolaclist
fre ay chat gpt-4o
feet DHCP request. toolsslist
SCP server Race request tools list
ira aes} WD running 1 tool
Se] BW curing tool. pypi_dvaniaads
Cray ae GO mP request tools/call oypr downloads
BB *CP server Nandle request: tools/call
@? pyps agent rua
QB chat aco sanoling
Bl CP request: sampling’ crestemessege
2 BY MCP client handle request: sampling/createnessage
foe eet] chat gnt-40
turning SELECT COAMTI¢] AS sue _doeslosds FROM Digquery-public-data... project + ‘pydantic ANO EXTRACT YEA@ FROM tinestal
Bigtvory JOD begis
BigQwery getQreryPosults
2 P request toolsslist
Megan — | 5 CP server haccile request’ tools/ist
oe ah ie as) ons
i
/
Pn el A
i IN - c an al H . aws at
es Mra sinol
ents Ley | a)
OCR text:
Thankyou
AIE Slidesat I'm at the Pydantic booth，ifyou have any questions, com and say hi.
World'sF- Graphite
World'sFair AEno iadd~Eair Microsoft aws smol?
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.