Improvado

Marketing data for AI agents: query ads, CRM and analytics data in your Improvado workspace.

Servidor MCP hospedado

npx add-mcp 'https://report.improvado.io/experimental/agent/api/mcp-customer/v1/invoke/'

Instala no Claude Code, Codex, Cursor e outros

Documentação

The Concept

Closed-Loop Marketing Execution

Your marketing team plans in decks. Launches in platform UIs. Measures in spreadsheets. Reports in yet another tool. Every handoff costs days. Context disappears between steps.

Closed-loop execution fixes that. Your AI agent runs the entire cycle — from hypothesis to better-performing campaign — across every channel, with every metric unified and every action governed. No handoffs. No context lost. And it gets smarter each cycle.

Ideate → Launch → Measure → Analyze → Report → Iterate ↻

Improvado MCP Powers the Loop

Model Context Protocol (MCP) is how AI agents talk to external tools. Anthropic built the standard. OpenAI and Google adopted it. Improvado's MCP server is the marketing-specific implementation — giving your AI agent read and write access to 1,400+ data sources through one governed connection. No engineering. No platform-by-platform integration.

Unified Data

46,000+ metrics harmonized via MCDM. "Cost" in Google Ads and "spend" in Meta resolve to the same field.

Governed Processes

Pre-built rules enforce naming, budgets, and KPI thresholds. Every query auditable. Every account change confirmed first.

Enterprise Security

Zero-trust proxy architecture. AI agents never see credentials. SOC 2, GDPR, HIPAA.

30 hrs

Per week saved in manual data aggregation and report assembly. Marketing teams shift from pulling data to acting on insights.

The Product

What Is Improvado MCP

MCP is the open standard for connecting AI to external tools. Anthropic built it. OpenAI and Google adopted it. 97 million SDK downloads a month — it's the way AI talks to the outside world now.

Improvado's MCP server exposes your entire marketing data platform — 1,400+ data sources, governed transformations, real-time analytics — to any AI agent through one secure interface. Ask a question, get real numbers. Not estimates. Not hallucinations. Numbers.

Technical Teams

Your Agent, Our MCP

Connect Claude, Cursor, Windsurf, or any MCP client directly. Full JSON-RPC 2.0 access. Your tools, our governed data layer.

Marketing Teams

Improvado AI Agent

Just chat. The AI Agent is the native MCP client — same 84 tools, no setup. Ask a question, get an answer with real numbers.

What Your Team Can Do — Mapped to the Loop

IdeateHistorical Insight Engine

"What creative format drove the lowest CPA for our fintech audience last quarter?" Real numbers from 1,400+ data sources. One question.

LaunchCampaign Execution

Build campaigns, configure A/B tests, set UTMs, push live — all through conversation. 1,400+ data sources behind the scenes.

MeasureAlways-On Governance

Budget pacing off? Caught hourly. Naming conventions broken? Flagged before it corrupts reports. Creative fatiguing? Detected and paused.

AnalyzeUnified Attribution

Cross-channel attribution where "cost" and "spend" are already the same field. CPA by segment, stat significance by variant.

ReportExecutive Reporting

CMO gets a strategic summary. Analysts get methodology. Finance gets budget impact. Same data, shaped for each audience.

IterateLearning Loop

"Variant B won by 34% — next test: same format, narrower audience, bigger budget." Agent proposes it. You approve. Cycle restarts.

↔

Same engine, two interfaces. Every tool above runs natively inside Improvado AI Agent. Your data team gets protocol-level control. Everyone else gets the same capability through conversation.

Getting Started

Connect Improvado MCP to Your AI Agent

Customer MCP uses OAuth — your existing Improvado credentials, no new tokens to manage. Eight methods depending on your environment. Most take under five minutes; Microsoft Copilot first needs a one-time OAuth application from Improvado.

Prerequisite

✓Access to your Improvado account (OAuth — no separate API key needed)

✓MCP endpoint: https://<your-domain>/experimental/agent/api/mcp-customer/v1/invoke

⚠️Every URL below uses <your-domain> — your Improvado domain. For a standard account use report.improvado.io, so the full endpoint URL is:

https://report.improvado.io/experimental/agent/api/mcp-customer/v1/invoke

If your organization uses a custom (whitelabel) domain, use that domain instead (e.g. acme.improvado.io). On a whitelabel domain, report.improvado.io will not authenticate you.

1

Claude Desktop App

Best for individual use. Connects in three clicks.

1

Open Connector Settings

In Claude Desktop, navigate to Customize → Connectors and click Add custom connector.

2

Paste the MCP URL & Connect

Enter the endpoint URL and click Connect. OAuth flow starts automatically.

https://<your-domain>/experimental/ agent/api/mcp-customer/v1/invoke

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

3

Authorize & Verify

Follow OAuth prompts with your Improvado credentials. Connector appears in active list once authorized.

2

Claude Code / Config File

Best for teams and programmatic access. Works in any project.

1

Option A: Add via CLI

claude mcp add --transport http improvado \ https://<your-domain>/experimental/\ agent/api/mcp-customer/v1/invoke/

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

2

Option B: Add to .mcp.json

Add to your project's config file:

{ "improvado": { "type": "http", "url": "https://<your-domain>/ experimental/agent/api/ mcp-customer/v1/invoke/" } }

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

3

Authenticate via OAuth

On first use, authorize with your Improvado account. All subsequent calls are automatic.

3

ChatGPT App

Best for marketers using ChatGPT. The same connection also exposes Improvado MCP inside OpenAI Codex and Codex CLI — no extra setup.

Improvado AI Agent in the ChatGPT App Store: find Apps in the left sidebar, then search Improvado ChatGPT New chat Search chats Library Apps Codex More Pinned Apps Chat with your favorite apps in ChatGPT Improva im Improvado AI Agent Work with marketing data ›

1

Open the ChatGPT App Store

In ChatGPT, click Apps in the left sidebar and search for Improvado.

2

Add the Improvado AI Agent app

Click the Improvado AI Agent result → Connect. OAuth flow starts automatically.

3

Authorize & use anywhere

Sign in with your Improvado credentials. Improvado is now available in any ChatGPT conversation, plus in Codex and Codex CLI through the same authorization.

4

Codex Plugin (Direct Install)

Best for engineers using OpenAI Codex directly. Adds Improvado as a Codex plugin available in both Codex and Codex CLI, including 20+ pre-built marketing skills (see list below).

1

Open Codex Plugins

In Codex, click Plugins in the left sidebar, then click the + button in the top-right to add a new plugin.

2

Enter the plugin settings

Fill the form with these values:

Source tekliner/improvado-agentic-frameworks-and-skills

Git ref master

Sparse pathsleave empty

3

Click Install

Codex pulls the plugin from the Improvado repository and adds it to your Added plugins list. The same Improvado tools are now available in both Codex and Codex CLI.

5

Gemini CLI

Best for engineers using Google's Gemini in the terminal. Note — the Gemini web app can't connect; Improvado MCP requires a local OAuth + HTTP MCP client (Gemini CLI or Antigravity).

1

Open the global config at ~/.gemini/settings.json and paste this into your mcpServers section. Reliable OAuth-scope configuration.

{ "mcpServers": { "improvado": { "url": "https://<your-domain>/experimental/agent/api/mcp-customer/v1/invoke/", "type": "http", "oauth": { "scopes": ["mcp:customer"] } } } }

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

2

Option B: Add via CLI

Run this in your terminal, then open ~/.gemini/settings.json and append the OAuth scopes to the new improvado entry.

gemini mcp add --transport http improvado \ https://<your-domain>/experimental/\ agent/api/mcp-customer/v1/invoke/

"oauth": { "scopes": ["mcp:customer"] }

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

3

Authorize

Inside an active Gemini CLI chat, run /mcp auth improvado to start the browser-based OAuth flow and connect your Improvado account.

6

Antigravity

Antigravity uses the same MCP server schema as Gemini CLI — add the Improvado entry to its MCP configuration, then authorize over OAuth.

1

Open the MCP configuration

In Antigravity, go to Settings → MCP servers → Manage MCP servers and open the JSON configuration for editing.

2

Add the improvado server

Paste the same block used for Gemini CLI:

{ "mcpServers": { "improvado": { "url": "https://<your-domain>/experimental/agent/api/mcp-customer/v1/invoke/", "type": "http", "oauth": { "scopes": ["mcp:customer"] } } } }

Replace <your-domain> with report.improvado.io (standard accounts) or your organization's whitelabel domain, e.g. acme.improvado.io.

3

Save & authorize

Save the configuration and reload the MCP server. On first connect, complete the browser-based OAuth flow to link your Improvado account. Verify by opening the MCP panel — improvado should appear as connected.

Before you connect Microsoft Copilot

The clients above register with Improvado automatically. Microsoft Copilot does not support dynamic client registration, so methods 7 and 8 need an OAuth application issued for your organization first.

1.Request an OAuth application. Contact your Customer Success Manager and ask for an Improvado MCP OAuth application for Microsoft Copilot. You will receive a Client ID and a Client Secret.

2.Allow-list your redirect URI. If your first authorization stops on an Improvado page saying the redirect URI is not registered or does not match, that is expected. Copy the full URL from the browser address bar and send it to your CSM. Once it is allow-listed, connect again.

7

Microsoft Copilot Studio

Best for teams building custom agents in Copilot Studio. Requires the OAuth application from your CSM (see above).

1

Open your agent

Go to Copilot Studio → Agents, then create a new agent or open an existing one.

2

Add an MCP tool

In the agent settings panel, click Add a tool, then + Add (top right) → Model Context Protocol (MCP). Improvado is not in the MCP tab catalog, so use + Add.Copilot Studio Add a tool panel with the + Add button highlighted

3

Fill in the MCP configuration

Server name: any name, e.g. Improvado MCP

Server description: e.g. MCP to interact with the Improvado platform

Server URL

https://report.improvado.io/experimental/agent/api/mcp-customer/v1/invoke/

Authentication: OAuth 2.0

Configuration type: Manual

Client ID / Client Secret: the values from your CSM

Authorization URL

https://report.improvado.io/api/dts/v2/oauth/o/authorize/?resource=https%3A%2F%2Freport.improvado.io%2Fexperimental%2Fagent%2Fapi%2Fmcp-customer%2Fv1%2Finvoke%2F&client_id=<client_id>&x=

Token URL and Refresh token URL

https://report.improvado.io/api/dts/v2/oauth/o/token/

Scopes: mcp:customer

In the authorization URL, replace <client_id> with your Client ID and keep the trailing &x= exactly as shown: Copilot appends its own parameters after it.

On a whitelabel domain, replace report.improvado.io with your domain in every URL, including the encoded copy inside resource=.

Copilot Studio Add MCP server form: server name, description, URL, OAuth 2.0, Manual Copilot Studio Add MCP server form: client ID, secret, authorization, token and refresh URLs, scopes

4

Create a connection

Click Add, then create a new connection to test the MCP. Sign in to Improvado if prompted. On the first attempt you will land on the redirect URI error: send that URL to your CSM (step 2 of Before you connect).

5

Reconnect & save

Once your CSM confirms the redirect URI is allow-listed, create the connection again and save the MCP configuration. Improvado MCP is now available to your agent.

8

Microsoft 365 Copilot (Desktop / Web)

Best for making Improvado available to Microsoft 365 Copilot agents. Requires access to Teams Developer Portal and the Microsoft 365 admin center, plus the OAuth application from your CSM.

1

In Teams Developer Portal, open Tools → OAuth client registration and fill in:

Registration name: any name, e.g. Improvado MCP

Base URL: https://report.improvado.io

Restrict usage by organization / by Teams app: per your policy

Client ID / Client Secret: the values from your CSM

Authorization endpoint

https://report.improvado.io/api/dts/v2/oauth/o/authorize/?resource=https%3A%2F%2Freport.improvado.io%2Fexperimental%2Fagent%2Fapi%2Fmcp-customer%2Fv1%2Finvoke%2F&client_id=<client_id>&x=

Token endpoint and Refresh endpoint

https://report.improvado.io/api/dts/v2/oauth/o/token/

Scope: mcp:customer

Enable PKCE: On

Client password authentication method: Request body parameters (default)

In the authorization URL, replace <client_id> with your Client ID and keep the trailing &x= exactly as shown: Copilot appends its own parameters after it.

On a whitelabel domain, replace report.improvado.io with your domain in every URL, including the encoded copy inside resource=.

Teams Developer Portal OAuth client registration form filled in for Improvado MCP

Save, then copy the OAuth client registration ID; you need it in the next step.

2

Create the connector

In the Microsoft 365 admin center, choose Created by your org → Create a new connector → Add and fill in:

Display name: e.g. Improvado MCP

MCP endpoint URL

https://report.improvado.io/experimental/agent/api/mcp-customer/v1/invoke/

Authentication type: OAuth 2.0

Reference ID: the OAuth client registration ID from step 1

3

Connect it to an agent

In Microsoft 365 Copilot, open the agent, go to Settings → Sources, find the Improvado connector and click Connect. Complete the Improvado sign-in. Improvado MCP is now available to the agent.

Marketing skills included with the plugin

The Codex plugin ships with 20+ pre-built skills covering common marketing workflows. Highlights:

CMO cross-channel dashboard

Executive marketing dashboard across spend, funnel, and attribution views.

Daily performance report

Operational daily performance console for multi-channel paid media.

Weekly creative performance

Creative dashboard with thumbnails, fatigue segments, Discovery API data.

Full marketing audit

Read-only paid-media audit: connection checks, per-channel review, synthesis.

Campaign launcher (OSS)

Open-source multi-channel campaign planning and launch workflow.

MDG create & edit

Marketing Data Governance rule creation, editing, testing, and alerts.

Discovery API

Live platform data access through Improvado Discovery API tools.

Recipe QA & rules

Recipe QA flow: topology discovery, QRDs, duplicate & discrepancy checks.

i

Once connected, the full 84 tools are available: query any metric, run governance checks, explore connectors, manage extracts, and more — all governed through Improvado's security layer.

Use Case

MCP + Knowledge Graph: Full Campaign in 1 Prompt

🧠

Knowledge Graph

ICP, brand guidelines, past experiments, budget rules, and channel history — all in persistent memory the agent reads before acting.

⚡

Improvado MCP

Live connection to 1,400+ data sources, email tools, and analytics APIs. The agent doesn't just read — it writes, launches, and adjusts.

Connect Improvado MCP to your AI agent and point it at your Knowledge Graph. A two-sentence prompt becomes a live multichannel campaign.

Prompt

"Run a new experiment called 'Marketing on Autopilot' — where Improvado agents plan, create, and launch marketing campaigns."

01

Read & Plan

Agent reads ICP, brand guidelines, and past experiments from the Knowledge Graph. Builds campaign plan: targeting, budget split, channel mix, KPIs. Asks for one validation.

→

02

Build in Parallel

Sub-agents fire simultaneously: banners generated (brand guidelines enforced), landing page built with all tracking pixels pre-wired, ad copy written, keyword lists created.

→

03

Launch & Enroll

Campaigns pushed live via MCP to Google Ads, Facebook, and email. Agent scores the lead database and auto-enrolls matched prospects into outreach sequences.

Google Ads

4 ad groups · responsive ads · sitelink extensions · new landing page linked

Facebook Ads

Enterprise broad + website retargeting · on-brand creatives · brand compliance enforced

Email + LinkedIn

LinkedIn invite · 2-email sequence · 50 pre-scored leads enrolled automatically

40–50

minutes prompt to live

3

channels launched in parallel

50

leads auto-enrolled

0

platform logins required

Who Benefits & How It Fits

Brand Roles That Benefit

One platform. Every marketing function accelerated.

CMO / VP Marketing

Cross-channel ROI in secondsBudget optimizationData-driven board reports

Marketing Analyst

Zero data wranglingAttribution modelingSelf-serve cross-platform queries

Marketing Ops Manager

Automated pipeline managementNaming convention governanceCampaign QA before launch

Performance Manager

Real-time budget pacingCreative fatigue detectionAutonomous optimization

Data / BI Engineer

No connector maintenanceStandard JSON-RPC protocolAny MCP client supported

Inside the Platform

Where MCP fits in Improvado's stack.

AI Agent + MCP

84 tools, any AI client

Dashboards & Reporting

15+ templates, native + BI

Marketing Data Governance

Pre-built rules for campaign health

Transform & Model

MCDM, 46K+ metrics unified

Extract & Load

1,400+ data sources, 20+ destinations

Enterprise Security

✓

Zero credential exposure. AI agents never see OAuth tokens or API keys. All credentials encrypted in a secure vault.

✓

Secure proxy architecture. MCP acts as authenticated intermediary. No direct API access from AI layer.

✓

Prompt injection immunity. Even if injection succeeds, nothing to exfiltrate — credentials never in agent context.

✓

Privacy-first LLM usage. Only metadata shared with AI models. Raw data never leaves your environment.

✓

SOC 2 Type II · GDPR · HIPAA. Enterprise compliance at every layer.

Appendix A

16 Use Cases

Marketing capabilities unlocked with Improvado MCP — common use cases across enterprise brands, agencies, and data teams.

Creative Analytics MCP as Agent Data Layer Data Quality Security & Infra Attribution

Creative Analytics · UC-01

Creative Image & Tag Data Direct to Data Warehouse

Top priority for enterprise teams: get raw creative image URLs, performance metrics, and AI-generated tags (archetype, tone, CTA type) from every ad platform directly into your data warehouse — no manual dashboard exports, no copy-paste between platforms.

discoveryRequestToolcreateExtractToolqueryDestinationTool

Raw URLs

not just creative IDs

10+ platforms

in a single pipeline

0

manual exports required

Creative Analytics · UC-02

Cross-Channel Creative Performance in Native Format

Many enterprises prefer raw, per-platform data over forced normalization due to very different ad units across platforms. MCP delivers channel-native creative performance — your own data warehouse models handle internal mapping, preserving existing architecture.

discoveryRequestTooldiscoveryListConnectorsToolclickhouseTool

Native

format per platform

250+

media properties

0

forced normalization

Creative Analytics · UC-03

Marketing Executive Dashboard with Custom Visualizations

Build interactive dashboards that combine cross-channel performance data with custom visualizations — heatmaps, funnel charts, trend analysis. CMO gets strategic KPIs, analysts get drill-down capability. One prompt, full dashboard.

clickhouseToolvisualizationToolheatmapVisualizationTool

Custom

visualizations

Real-time

data refresh

Executive

-ready output

MCP as Agent Data Layer · UC-04

Native MCP Endpoint for Your Internal Agents

Building multi-tenant AI agents in-house? MCP provides the marketing data retrieval endpoint those agents need — no Improvado agent required. Your agents call MCP tools through your AI gateway after a single security review.

getCurrentWorkspaceToolswitchWorkspaceToolmcpListToolsTool

1 review

unlocks all 84 tools

84 tools

MCP tool catalog

0

manual data pulls

MCP as Agent Data Layer · UC-05

Live Spend Data for Budget Management Automation

Use MCP data for budget management. Discovery API delivers today's spend vs. budget from all active channels without ETL lag — giving your internal budget agents real-time data to act on, not yesterday's batch report.

discoveryRequestToolclickhouseTooldiscoveryListConnectorsTool

60 min

pacing cycle vs 24h

250+

channels live

Real-time

not batch data

MCP as Agent Data Layer · UC-06

Self-Serve Marketing Queries — No Data Engineer Bottleneck

When only a few data engineers have platform access, marketers pull data from native platform UIs. MCP exposes unified marketing data to your agent layer — enabling natural-language marketing queries without creating a ticket to the data team.

clickhouseToolvisualizationToolfindChat

1→many

analyst access

0

tickets to data team

Natural

language queries

Data Extraction · UC-07

Extract Data from 1,400+ data sources

Connect to any marketing data source — ad platforms, analytics tools, CRM systems, email providers. Improvado's 1,400+ data sources handle authentication, pagination, rate limiting, and schema mapping automatically. One API, unified access.

discoveryListConnectorsToolcreateExtractToolcreateConnectionTool

1,400+

data sources

46K+

unified fields

0

connector maintenance

Data Quality · UC-08

Clean Marketing Data Inputs for ML Models

Enterprises deploying ML models at scale need clean inputs. Stale or corrupted marketing channel data silently degrades recommendation quality and conversion rates. MCP's MDG layer monitors data freshness and schema integrity at the connector level before it reaches the model pipeline.

createMdgViewToolgetMdgRuleLibraryExampleToollistAllExtractsTool

1000s

models need clean inputs

Hourly

freshness checks

0

silent degradation

Data Quality · UC-09

Campaign Naming Governance for Data Warehouse Join Integrity

Malformed UTMs and inconsistent naming break attribution lookups when you join Improvado data with internal data warehouse tables. MDG enforces naming at ingestion — every ad campaign record reliably joinable to your conversion event data.

createMdgViewToolgetMdgRuleLibraryExampleTooltestMdgNotificationTool

MDG

rules enforced

Ingest

-point enforcement

0

broken DW joins

Data Quality · UC-10

Quality Gate for New Ad Network Onboarding

Actively adding new ad networks? Each new connector introduces schema formats that can corrupt existing attribution joins. MCP validates new source data against your taxonomy before it lands in your data warehouse.

createConnectionToolcreateExtractToolcreateMdgViewTool

1,400+

data sources available

Auto

schema validation

0

schema breaks

Security & Infrastructure · UC-11

MCP Through Your AI Gateway — Single Security Review

Enterprise security teams require review for every new AI integration. Improvado MCP's workspace-scoped OAuth session model is designed to pass security review cleanly — no direct IAM, no credential exposure. One review unlocks all 84 tools, not one per platform.

getCurrentWorkspaceToolgetConnectionsToolswitchWorkspaceTool

1 review

for all 84 tools

OAuth

not direct IAM

0

credentials exposed

Security & Infrastructure · UC-12

Native Data Warehouse Access Avoids IAM Security Reviews

Many enterprises route destinations through their data warehouse to avoid direct IAM access triggering multiple security reviews. MCP's queryDestinationTool queries your BigQuery/Snowflake natively — no direct IAM, no new credentials, no additional security review.

queryDestinationToolgetDestinationConnectionsToolclickhouseTool

BQ-native

no IAM required

0

additional reviews

Matches

enterprise DW patterns

Operational Efficiency · UC-13

Discovery API Reduces Unnecessary Extraction Volume

Many enterprises don't use their full annual row contract. Discovery API enables real-time ad-hoc queries against live platform APIs — reducing scheduled extraction volume while providing fresher data for time-sensitive queries. Fewer rows consumed, better data quality.

discoveryRequestToolpauseExtractToollistAllExtractsTool

50%+

row contract unused

Discovery

for live queries

Fewer

scheduled extractions

Attribution & Measurement · UC-14

Paid Spend → Conversion Attribution Loop

Query paid media spend (Improvado MCDM) and join with conversion events from your data warehouse in one MCP session. This is the core measurement problem every performance marketing team needs to solve — revenue traced back to campaign impression.

clickhouseToolqueryDestinationToolvisualizationTool

2 layers

1 MCP session

Impression

→ revenue closed

0

data exports needed

Attribution & Measurement · UC-15

90-Day Multi-Touch Attribution for High-Consideration Products

High-consideration purchase decisions take 30–90 days. A prospect sees an ad in January and converts in March. MCP's 90+ day MCDM history enables multi-touch attribution across the full consideration window — connecting the January impression to the March conversion.

clickhouseToolqueryDestinationToolcreateRecipeTool

90-day

lookback window

Multi-touch

not last-click

46K+

unified metrics

Attribution & Measurement · UC-16

Revenue Yield Optimization by Channel & Creative

High-value conversions drive revenue. MCP enables revenue-per-impression analysis by channel, daypart, creative archetype, and audience segment — shifting optimization from click rate to conversion rate, the metric that actually drives revenue.

clickhouseToolqueryDestinationToolvisualizationTool

High LTV

per conversion

Conversion

rate not click rate

Revenue

yield per impression

Appendix B — For Engineering, Data, & InfoSec Teams

Architectural Highlights

4 Improvado MCP design decisions that map directly to enterprise infrastructure constraints — AI gateway integration, data warehouse destination, security review process, and intraday data needs.

Architecture Highlight · AH-01

Workspace-Scoped Sessions Ready for Your AI Gateway

Common question: "Do we need our networking team involved?" Yes — and MCP is built for it. Every session is a workspace-scoped OAuth token (15-min Redis TTL) that passes through an AI gateway cleanly. Multi-tenant agent infrastructure maps 1:1 — each agent gets its own scoped session, preventing cross-tenant leakage by architecture, not policy.

15 min

session TTL, Redis-backed

0

cross-tenant leakage

1:1

agent-to-workspace scope

Architecture Highlight · AH-02

queryDestinationTool: Native Data Warehouse, No IAM Review

Many enterprises use BigQuery/Snowflake as primary destination — and route data through their warehouse specifically to avoid IAM-triggered security reviews. MCP's queryDestinationTool queries your own warehouse tables natively in the same session as MCDM data — enabling the spend-to-conversion join, with zero new IAM access required.

BQ-native

no IAM triggered

0

additional reviews

2 sources

1 MCP session

Architecture Highlight · AH-03

Zero Credential Exposure — Designed for Enterprise Security Review

Security teams will ask: "Can the AI agent see our platform credentials?" The answer is architecturally no. Your Google Ads, Meta, and other OAuth tokens are stored AES-256 encrypted in Improvado's PostgreSQL vault. The MCP server calls platform APIs internally — credentials never appear in any tool response, LLM context window, or network log. SOC 2, GDPR, CCPA compliance is structural, not a configuration option.

0

credentials in agent context

SOC 2

Type II certified

AES-256

at-rest encryption

Architecture Highlight · AH-04

Three Data Layers in One Session: Live + Historical + DW

Common question: "Does MCP retrieve only, or does it analyze?" It retrieves from three layers simultaneously — Discovery API for real-time platform data (no ETL), ClickHouse MCDM for 90+ days of harmonized history, and queryDestinationTool for your own data warehouse. Your agents then own the analysis layer — which is exactly what most enterprises want.

<60s

live platform data

90+ days

MCDM history

3 sources

1 agent session

Appendix C — Public Endpoint

MCP Tool Reference

Complete tool inventory for Improvado MCP public endpoint. Every tool is accessible via JSON-RPC 2.0 after OAuth authentication — no per-tool setup required. Tool categories below match the use cases in Appendix A.

Data Query & Visualization 5 tools

clickhouseToolExecute SQL against ClickHouse warehouse — customer workspace scope

queryDestinationToolQuery downstream destinations directly (BigQuery, Snowflake, Redshift)

visualizationToolGenerate charts — line, bar, area, pie, scatter, table

heatmapVisualizationToolCreate heatmaps for time-based metric analysis

sendEmailToolSend reports and visualizations via email

Discovery API — Live Platform Access 5 tools

discoveryListDataSourcesToolList all data sources available in the workspace

discoveryListConnectorsToolList connectors/integrations by data source name

discoveryListAccountsToolList ad accounts within a specific connector

discoveryRequestToolExecute live query against any of 1,400+ data sources (Google Ads GAQL, Meta, etc.)

discoveryWebSearchToolSearch documentation and support resources for a data source

Connections & Accounts 3 tools

getConnectionsToolGet all connections for a datasource in current workspace

getConnectionAccountsToolGet ad accounts associated with a specific connection

getDestinationConnectionsToolList destination connections (where data lands)

ETL — Extract Management 15 tools

createExtractToolCreate a new scheduled data extraction from a source

updateExtractToolUpdate extract schedule, accounts, or field selection

getExtractByIdToolGet details of a specific extract by ID

getExtractTemplatesToolList available extract templates

getExtractTemplateByIdToolGet a specific extract template by ID

listAllExtractsToolList all extracts in the workspace with status; filter by data table, account, status or title

pauseExtractToolPause a running extract (stops scheduled pulls)

unpauseExtractToolResume a paused extract

discoverFileFieldsToolDiscover available fields for a file-based data source

bulkRunExtractsToolRun many extracts at once, selected by ID or by data table

bulkPauseExtractsToolPause many extracts at once

bulkUnpauseExtractsToolResume many paused extracts at once

bulkEditExtractsToolChange settings for many extracts in one request (replace, add, remove or update values)

getExtractionBulkTaskToolCheck the status and result of a bulk action

ETL — Load Management 8 tools

createLoadToolCreate a new load to push data to a destination

updateLoadToolUpdate load destination or schedule settings

getLoadToolGet details of a specific load by ID

getLoadsToolList loads with filtering options

listLoadsToolList all loads in the workspace

deleteLoadToolDelete a load permanently

pauseLoadToolPause a running load

unpauseLoadToolResume a paused load

ETL — Run Management 4 tools

runLoadToolManually trigger a load run immediately

getPipelineRunsToolGet run history for a pipeline

getPipelineRunResultToolGet detailed results of a specific pipeline run

generateFlatDataSampleToolGenerate a data sample from a flat file source

Pipelines 6 tools

createCustomPipelineToolCreate a custom SQL/Python transformation pipeline

updateCustomPipelineToolUpdate pipeline logic or schedule

deleteCustomPipelineToolDelete a pipeline

listCustomPipelinesToolList all pipelines in workspace

executeCustomPipelineToolExecute a pipeline immediately

runCustomPipelineCodeToolRun pipeline code with custom parameters

Marketing Data Governance / MDG 7 tools

createMdgViewToolCreate a governance rule/view (naming, budget, classification)

getMdgRuleLibraryExampleToolGet examples from the pre-built MDG rule library

testMdgNotificationToolTest MDG alert and notification configuration

previewEmailImportAddressToolPreview email import address for data ingestion

uploadFlatDataFileToolUpload CSV or Excel file for processing

uploadNewFileToExtractToolUpload a new file version to an existing extract

createFlatDataExtractToolCreate an extract pipeline from flat file data

AI Recipes / Notebooks 13 tools

createRecipeToolCreate a new composable AI analysis notebook

updateRecipeToolUpdate recipe settings, cells, or parameters

getRecipeToolRetrieve a recipe by ID

getRecipeTemplatesToolList available recipe templates (attribution, reporting, etc.)

createRecipeStepToolAdd a SQL or Python step to an existing recipe

getRecipeStepToolGet a specific step within a recipe

listRecipesToolList all recipes in the current workspace

activateRecipeToolActivate and run a recipe

nameRecipeToolSet or update the display name of a recipe

createLesIntegrationChatToolCreate a LES integration chat session for guided setup

createAgentAutomationToolSchedule the agent to run a prompt once or on a recurring schedule (an agent automation)

getAgentAutomationsToolList agent automations with their latest run status

deleteAgentAutomationToolDelete an agent automation and stop its future runs

Dashboards 3 tools

getDashboardToolRetrieve a dashboard by ID

listDashboardsToolList all dashboards in the workspace

getCustomDashboardsAndDecksPromptToolGet context prompt for custom dashboards and decks

Documents · Chat · Knowledge · Data Tables 11 tools

createDocumentCreate a new document in the workspace

editDocumentEdit content of an existing document

getDocumentRetrieve a document by ID

deleteDocumentDelete a document permanently

listDocumentsList all documents in the workspace

findChatSearch past AI conversations by query

getChatGet full conversation context and history by ID

getManagedConnectionCredentialsToolGet connection credential metadata (IDs/status — never raw secrets)

documentationToolQuery Improvado product documentation

retrieveDocumentationToolRetrieve specific documentation files by path

listDataTablesToolList all data tables available in current workspace

Workspace · MCP Protocol · Connection Setup 9 tools

getCurrentWorkspaceToolGet current workspace context — agency, workspace ID, name

switchWorkspaceToolSwitch to a different workspace within allowed agencies

getWorkspacesToolList all workspaces available to the authenticated user

mcpListToolsToolDynamically list all available MCP tools at runtime

mcpCallToolToolCall any MCP tool by name (meta-wrapper for agents)

mcpListResourcesToolList MCP resources available in current context

mcpReadResourceToolRead a specific MCP resource by URI

createConnectionToolCreate a new data source connection in the workspace

customConnectorFormToolAccess custom connector configuration form

Questions? The Improvado team is always happy to help. Send us an email, contact your Customer Success Manager, or raise a request in the Improvado Service Desk.