NEWUnified secure data

Give OpenAI Codex a real
data layer

Connect Codex to Improvado and your coding assistant gets clean spend, attribution, and CAC data - plus the shared memory your whole team already uses.

Blended CAC by channel · last 30 dayslive

Which channels are over CAC?

Querying your governed layer via Improvado MCP

Codex

CAC vs $XXX targetblended $XXX

Google Ads$XXX

Meta Ads$XXX

TikTok Ads$XXX

TikTok CAC is $XXX - XX.X% over your $XXX target. The only channel flagged.

In [1]:

from improvado import mcp
df = mcp.query("spend, cac by channel", window="last_30d")
df.plot.bar(x="channel", y="spend")

Out[1]:

spend by channel · last_30d

$XXXK$XXXK$XXXK$0

$XXXKgoogle_ads

$XXXKmeta_ads

$XXXKtiktok_ads

In [2]:

df[df.cac > mcp.memory.recall("Q3 CAC target") * 1.10] # target $XXX

Out[2]:

tiktok_ads CAC $XXX is +XX.X% over the $XXX target - only channel flagged

Scheduled jobs6 active

job	schedule	last run	status
attribution_export.py	Daily 06:00	06:02 UTC	success
cac_alert.py	Every 4h	2h ago	success
weekly_board_pack.py	Mon 07:00	3 days ago	success
anomaly_scan.py	Hourly	18 min ago	retried x1
creative_fatigue_scan.py	Daily 05:30	5h ago	success
budget_pacing_check.py	Every 6h	1h ago	success

next: cac_alert.py · run now to check

cac_alert.py blended CAC $XXX, -$XX vs target - no alert fired

run idrun_8f21c40

exit code0

cron*/4 * * * (every 4h)

duration412 ms

retries0 of 2

thresholdCAC > $XXX +XX%

alert routing#growth · growth team

last triggernone (steady-state)

Codex authors each script against the MCP schema; Improvado runs and monitors the schedule.

Trusted by marketing teams at All customer stories →

How it works

From question to scheduled report in three steps

Connect once, ask in plain English, and put the answer on autopilot - no setup project and no query language to learn.

Connect

Connect Improvado to Codex once - set up at onboarding, nothing to install.

PII masked · SOC 2 Type II · EU residency

Set up once at onboarding - nothing for your team to install or maintain.

Ask anything

Ask about spend, CAC, LTV, or attribution across every connected source and get the answer in seconds - in plain numbers, not a query language.

Is anything burning budget this month?

$XXX (X% under target) ·

$XXX (+X%) · TikTok flagged at $XXX, XX.X% over.

Which one should I pause first?

TikTok - pausing saves ~$XXK/mo at current CAC; Search and Meta stay on plan.

412 ms median answer · every number source-cited · saved to shared memory

Schedule it

Turn any answer into a report that runs on a schedule and lands where your team needs it - nobody has to remember to pull it.

Every morning · 6:00 AMnext in 3h

CAC report delivered · flags anything over target

7 of 7 delivered2 mornings flagged

Runs on a schedule and saves the results.

See the technical setupThe exact commands, queries & scheduling - for your engineers and AI agents

1Show setup command2See the notebook cell3See the scheduled job

# 1 - Show setup command
{ "mcpServers": {
    "improvado": { "url": "https://mcp.improvado.io/v1" }
} }
✓ oauth: authenticated as growth@acme.com
✓ scopes: read:spend · read:crm · write:memory
# 2 - See the notebook cell
In [1]: mcp.query("blended CAC by channel, 30d")
Out[1]:  google_ads   $XXX   -7% vs target
         meta_ads     $XXX   +3% vs target
         tiktok_ads   $XXX   +23.7% flagged
# 3 - See the scheduled job
cron: "0 6 * * *"  cac_alert.py
today      success · 412 ms · 3 flags → knowledge graph
yesterday  success · 398 ms · 0 flags
Mon        success · 405 ms · 1 flag

What growth teams run with it

Pick a workflow, see the live board

Live boards from your connected data - pick a workflow to see what your team gets.

Attribution · blended ROAS

3.8×+0.4 vs last moblended ROAS
last 30 days

Google Ads

4.1×

Meta Ads

3.4×

TikTok Ads

3.1×

Best channelGoogle Ads · 4.1×

Trend 30d+0.4×

CAC monitor

$XXX-$XX vs targetblended CAC
last 30 days

Google Ads

$XXX

Meta Ads

$XXX

TikTok Ads

$XXX

Vs target-$XX blended

WatchTikTok Ads · $XXX

Anomaly scan

3flagged · 24hanomaly scan
last 24 hours

TikTok spend

+XX%

Meta CTR

-XX%

Google CAC

+X%

Flagged3 of 52 accounts

Next scan42 min

Scheduled reports

40 failedscheduled jobs
last 7 days

Weekly ROAS report

7/7

Daily CAC alert

42/42

Hourly anomaly scan

165/168

Success rate214 / 217 runs

Next runDaily CAC alert · 2h

See the queries behind these boardsThe exact query each workflow runs - for your engineers and AI agents

> query: metric=roas channel=all window=30d

> query: metric=cac channel=all window=30d

> query: scan=anomaly window=24h threshold=2.5_sigma

> query: jobs=scheduled window=7d status=all

Why Codex + Improvado

Clean marketing data, ready the moment you ask

Codex is only as good as the data it can reach. Improvado hands it the same clean, governed data your whole team already trusts.

1,000+ sources · one connection

All your channels in one ask

Every connected source in seconds - one clean set of numbers, nothing to wire up.

214 facts · shared across agents

Shared memory with your team

The same shared memory Claude and your team use - context carries across every tool.

Codex reads through one governed endpoint - scoped per seat, every call logged, and no platform credential is ever stored on your side.

Enterprise security

SOC 2 Type II, GDPR-ready, role-based access and audit logs - no credentials anywhere.

Runs anywhere Codex does

Works wherever your team works

Use Codex however your team already does - the connection follows wherever it runs.

Every channel · one format

One clean format

Every channel arrives in one format - what works for Google Ads works for Meta and TikTok too.

Live · 52 accounts, updated 41s ago

Always up to date

Metrics are live - what Codex reads is exactly what your dashboards show.

See the technical detailsHow Codex reads each capability - for your engineers and AI agents

All your channels in one ask

query.mcp()

improvado query --sources google_ads,meta,tiktok { "spend": 184230.55, "source": "unified_marketing" }

Shared memory with your team

mem.sync()

-> graph: 214 nodes · shared across agents

Enterprise security

auth.scope()

-> scope=read:campaigns · 0 creds issued

Works wherever your team works

mcp.connect()

-> session open · kernel py3.11

One clean format

schema.map()

spend↦google_ads.cost

spend↦meta.spend

Always up to date

stream.live()

-> lag 41s · 52 accounts current

From question to answer

Minutes to your first answer, not a data project

Connect once and Codex queries your live marketing data the same day - no data project, no integration backlog, no waiting on your data team.

Weeks become minutes

Ask on day one. The data-engineering project that used to sit between you and an answer is already done.

One connection, every channel

Every ad platform, your CRM, and your warehouse arrive as one clean set of numbers - not 500 exports to reconcile.

Numbers you can trust

Governed access and one clean, consistent data source mean Codex's answer matches what your BI team would hand you.

See the full technical comparisonHow this compares to wiring up each platform yourself

Dimension	Raw platform APIs	With Improvado MCP
Auth	-One credential per platform	+One OAuth connection
Schema	-Drifts per source, per version	+One normalized schema
Memory	-None - re-explain context every session	+Shared knowledge graph
Credentials	-API keys pasted into prompts	+Zero credentials leave the endpoint
Rate limits	-Per-API, hit blind	+Governed layer, one budget
Time to first query	-Weeks of glue code	+Minutes, one config block
Dependency	-N direct integrations you maintain and control	+One governed endpoint - a single vendor dependency, by design

# net: 7 dimensions, 1 governed endpoint

Architecture

One governed layer between Codex and your entire stack

Switch AI tools whenever you want. Your data, your access rules, and everything your team has taught the system stay exactly where they are - nothing to reconnect, nothing relearned.

Switch agents. Keep your data. Keep your memory

ConnectedImprovado data layerX,XXX sources · one clean format

set up once - every agent on the team reads the same clean data

Shared memory3 nodes - synced 2m agoQ3 CAC targetpipeline reviewspend anomaliespersists across every switch

OpenAI Codex reads your live data directly - no credentials leave your workspace.

For engineers & AI agentsThe MCP contract, endpoints, and how memory persists across agents

1One connection, declared once2Scopes - role-based, read-only by default3Shared memory - survives an agent switch

# 1 - One connection, declared once
{
  "mcpServers": {
    "improvado": {
      "url": "https://mcp.improvado.io/v1",
      "auth": "oauth"
    }
  }
}
# 2 - Scopes - role-based, read-only by default
# read:spend
# read:crm
# read:attribution
# write:memory
# 3 - Shared memory - survives an agent switch
persisted in the governed knowledge graph · versioned · every agent (Claude, Codex, custom) reads the same nodes

Enterprise-grade security

Agent access your security team will actually approve

Why it matters

Security compliance is where most AI coding assistants slow down. It's where we differentiate.

Get your demo

SOC 2 Type II · HIPAA · GDPR/CCPA · 6/6 controls green, audited quarterly · every query logged, zero raw credentials.

For engineers & AI agentsThe live compliance audit - every control, cycle date, and scope

audit@improvado:~$ compliance --check --workspace=prod --verbose

soc2-type2PASSaudited 2026-04-11 · next cycle 2027-04-11

hipaa-baaPASS47 healthcare controls verified

role-based-aclACTIVE212 scoped policies enforced

audit-trailLOGGINGXXX% query coverage, 13-month retention

sso-samlACTIVEOkta, Azure AD, Google Workspace

gdpr-ccpaPASSEU residency: eu-central-1 (Frankfurt)

6/6 controls greenOKlast full scan 41 min ago

SOC 2 Type II

Audited annually

HIPAA

Healthcare-grade

Role-based access

Per user & workspace

Full audit trail

Every query logged

SSO & SAML

Enterprise identity

GDPR & CCPA

EU data residency

Customer story

eicoff_verbatim.log

Improvado powers our data views every single day. It's a pulse check on business performance that enables our clients to make smarter, strategic decisions on budgeting, sales forecasting, and assess marketing's impact on their businesses.

Bill Urciuoli

SVP, Innovation · Eicoff

Read the full story

$XXXK/yr

Savings in staff costs

15x

Faster client onboarding

<1 wk

Time to value

FAQ

Common questions

Q01Does Codex support MCP?

Yes - Codex works with MCP servers. Add the Improvado endpoint to your Codex configuration and authenticate; every connected source becomes available in a notebook cell or a script.

Q02Can Codex and Claude share context?

Yes. Both read and write the same Improvado knowledge graph, so a target set in one session is recalled in the other - agents and humans stay consistent.

Q03Can Codex schedule recurring jobs against the data?

Yes. A script Codex writes against the MCP layer can be scheduled to run on a cron - exports, alerts, and reports keep running without a manual pull, and failures surface back in the same workspace.

Q04Is my data safe?

Yes. Agents connect through Improvado's MCP layer with scoped, role-based access and a full audit trail. No raw database credentials are ever shared with an agent, and every query is logged. The platform is SOC 2 Type II audited and GDPR-ready.

Q05What is MCP?

MCP (Model Context Protocol) is the open standard AI agents use to reach external tools and data. Improvado exposes your entire connected marketing stack - 1,000+ sources, unified schema, knowledge-graph memory - through one MCP endpoint.

Q06How long does setup take?

Minutes. If your sources are already connected to Improvado, you add one MCP config to your agent and authenticate with OAuth. No pipelines to build, no schemas to map.

$still have questions? talk to an engineer

Any agent, same data

Built for whichever agent your team already runs

One connection, one clean format, one shared memory - swap the agent, keep everything.

ClaudeChat-first agent workflows View page →

OpenAI CodexReports and scheduling You're here

Your custom agentBuild on the same clean data View page →

See it on your own data

Give Codex a real measurement foundation

Bring your own stack - we'll show your pipeline, attribution, and Codex querying it live, in twenty minutes.

$ codex book-session --duration=20m --stack=your-data

Book a working session

A solutions engineer maps your stack and shows Codex querying your own pipeline - no slideware.

[20m demo] [your data] [live in 14d]

1,000+

Connectors

90 hrs

Saved per week

XX%

Less overspend

XX%

Faster launches

Platform-wide results across Improvado customers - not specific to any single agent.

Give OpenAI Codex a realdata layer

From question to scheduled report in three steps

Connect

Ask anything

Schedule it

Pick a workflow, see the live board

Clean marketing data, ready the moment you ask

All your channels in one ask

Shared memory with your team

Enterprise security

Works wherever your team works

One clean format

Always up to date

Minutes to your first answer, not a data project

One governed layer between Codex and your entire stack

Switch agents. Keep your data. Keep your memory

Agent access your security team will actually approve

Common questions

Built for whichever agent your team already runs

Give Codex a real measurement foundation

Give OpenAI Codex a real
data layer