BlueAlpha is now on

BlueAlpha MCP

The BlueAlpha MCP server connects Claude Desktop, Codex, and other MCP-compatible clients to your BlueAlpha workspace — your Marketing Mix Model, your Google Ads account, and your Meta Ads account — as a single authenticated tool surface.





Open your AI assistant, ask a question, get a grounded answer with methodology attached.

Prerequisites

Before you install, make sure you have:

An MCP-compatible client installed on your machine. We currently support:

  • Claude Desktop

  • Codex

Connected accounts inside your workspace for the data you want to query:

  • Google Ads account with read/query access (optional - needed for Google Ads tools)

  • A fitted Meridian MMM (optional - needed for MMM tools)

You don't need a Python environment, Docker, or any local model files. The MCP server is hosted by BlueAlpha.

Installation

There are two ways to connect BlueAlpha to your AI assistant. Pick the one that matches how you work.

01

Open Claude Desktop

02

Go to Settings → Connectors

03

Add a custom connector. Name it “BlueAlpha MCP”

04

Add this URL: https://mcp.bluealpha.ai/mcp and click Connect & Authenticate.

05

You’re Done!

Validate your setup

Open your client and try one of these prompts. If you see structured output from your account, you're live.

"Check my Google Ads connection" 


"List the MMM models in my workspace" 


"Show me my Meta account structure for the last 30 days"

What's in the surface

The BlueAlpha MCP exposes three capability groups today through a single authenticated connection. Every response is structured, carries its methodology forward, and (for MMM outputs) returns uncertainty alongside the point estimate.

Marketing Mix Model (Meridian)

Function

What it does

Example questions

list_models

/

load_model

/

upload_model

Model file management

"What models do I have saved?" / "Load the US Q3 model"

get_model_summary

Metadata: channels, time range, data coverage

"What's in this model - which channels, what date range?"

get_model_settings

MCMC config, chains, tuning, training details

"What were the training settings for this model?"

get_channel_roi

Posterior ROI by channel (median, mean, credible intervals)

"What's the ROI of each paid channel with uncertainty bounds?"

get_marginal_roi

mROI — return on the next dollar

"If I had $1 more to spend, where should it go?"

get_contribution_breakdown

% of revenue attributable to each channel

"How much of total revenue did paid media drive vs baseline?"

get_weekly_contributions

Channel contribution over time

"Show me how Meta's contribution trended week-over-week"

get_saturation_curves

Hill saturation parameters per channel

"Is YouTube saturated at current spend levels?"

get_adstock_parameters

Carry-over decay rates

"How long does a TV impression keep working?"

get_channel_priors

Prior distributions used in the model

"What priors did we set going in?"

get_prior_posterior_comparison

Priors vs what the model actually learned

"Where did the data update our beliefs the most?"

simulate_budget_reallocation

Project revenue impact of a proposed budget shift

"If I move 20% from Meta to TikTok, what happens to revenue?"

Function

What it does

Example questions

list_models

/

load_model

/

upload_model

Model file management

"What models do I have saved?" / "Load the US Q3 model"

get_model_summary

Metadata: channels, time range, data coverage

"What's in this model - which channels, what date range?"

get_model_settings

MCMC config, chains, tuning, training details

"What were the training settings for this model?"

get_channel_roi

Posterior ROI by channel (median, mean, credible intervals)

"What's the ROI of each paid channel with uncertainty bounds?"

get_marginal_roi

mROI — return on the next dollar

"If I had $1 more to spend, where should it go?"

get_contribution_breakdown

% of revenue attributable to each channel

"How much of total revenue did paid media drive vs baseline?"

get_weekly_contributions

Channel contribution over time

"Show me how Meta's contribution trended week-over-week"

get_saturation_curves

Hill saturation parameters per channel

"Is YouTube saturated at current spend levels?"

get_adstock_parameters

Carry-over decay rates

"How long does a TV impression keep working?"

get_channel_priors

Prior distributions used in the model

"What priors did we set going in?"

get_prior_posterior_comparison

Priors vs what the model actually learned

"Where did the data update our beliefs the most?"

simulate_budget_reallocation

Project revenue impact of a proposed budget shift

"If I move 20% from Meta to TikTok, what happens to revenue?"

Google Ads

Function

What it does

Example questions

check_connection_status

Verifies your Google Ads account is connected

"Is my Google Ads account connected?"

list_accessible_customers

Lists all Google Ads accounts you can access

"Which Google Ads accounts do I have access to?"

execute_query

/

execute_query_stream

Runs any GAQL query with pagination/streaming

"Pull me last 30 days of campaign spend, clicks, and conversions"

get_field_metadata

/

validate_query_fields

GAQL schema helpers

"Can I select these fields together in one query?"

audit_campaign_structure

Full campaign structure & settings audit

"Audit my account - what's structurally wrong?"

analyze_audiences

Audience performance + efficiency scoring

"Which audiences are pulling their weight vs wasting spend?"

analyze_creatives

Score and classify ad creatives

"Which of my ads are top performers vs underperformers?"

detect_creative_fatigue

Last 7d vs prior 7d creative comparison

"Which creatives are fatiguing?"

diagnose_underspend

5-layer diagnostic for why campaigns aren't spending

"Why isn't my Performance Max campaign hitting budget?"

analyze_budget_reallocation

Efficiency-based reallocation analysis

"Where should I shift budget to improve overall ROAS?"

get_recommendations

Google's own optimization recommendations

"What is Google suggesting I do?"

generate_keyword_ideas

Pulls from Google's Keyword Planner

"Give me keyword ideas for 'marketing mix modeling'"

suggest_geo_targets

Geo target constants by name/address

"Find the geo target IDs for top 10 US metros"

monitor_incrementality_test

Integrity checks for incrementality tests

"Is my current geo test still clean, or has something leaked?"

Function

What it does

Example questions

check_connection_status

Verifies your Google Ads account is connected

"Is my Google Ads account connected?"

list_accessible_customers

Lists all Google Ads accounts you can access

"Which Google Ads accounts do I have access to?"

execute_query

/

execute_query_stream

Runs any GAQL query with pagination/streaming

"Pull me last 30 days of campaign spend, clicks, and conversions"

get_field_metadata

/

validate_query_fields

GAQL schema helpers

"Can I select these fields together in one query?"

audit_campaign_structure

Full campaign structure & settings audit

"Audit my account - what's structurally wrong?"

analyze_audiences

Audience performance + efficiency scoring

"Which audiences are pulling their weight vs wasting spend?"

analyze_creatives

Score and classify ad creatives

"Which of my ads are top performers vs underperformers?"

detect_creative_fatigue

Last 7d vs prior 7d creative comparison

"Which creatives are fatiguing?"

diagnose_underspend

5-layer diagnostic for why campaigns aren't spending

"Why isn't my Performance Max campaign hitting budget?"

analyze_budget_reallocation

Efficiency-based reallocation analysis

"Where should I shift budget to improve overall ROAS?"

get_recommendations

Google's own optimization recommendations

"What is Google suggesting I do?"

generate_keyword_ideas

Pulls from Google's Keyword Planner

"Give me keyword ideas for 'marketing mix modeling'"

suggest_geo_targets

Geo target constants by name/address

"Find the geo target IDs for top 10 US metros"

monitor_incrementality_test

Integrity checks for incrementality tests

"Is my current geo test still clean, or has something leaked?"

Best practices

01

Ask outcome questions, not tool questions

The MCP routes your question to the right tool automatically. You'll get sharper answers by framing the outcome you want, not the analysis you think you need.
Weaker prompt: "Run a GAQL query on my campaign performance for the last 30 days."
Stronger prompt: "Which of my Google Ads campaigns are underperforming, and what should I do about them?"

02

Anchor every question in your business context

The more context you put in the prompt, the tighter the recommendation. Tell the model your spend level, goal metric, target CAC or ROAS, and time horizon.
Weaker prompt: "Where should I reallocate budget?"
Stronger prompt: "We spend $130K/week across paid. We want to grow conversions 10% next month without changing the budget. Where should I reallocate?"

03

Read the methodology before you act

Every answer carries the methodology forward - which model, which time range, which posterior samples, which confidence interval. Before you execute a budget move, glance at the methodology footer. It tells you how much weight to put on the answer.

04

Chain questions within a session

The MCP keeps context across your conversation. Start broad, drill in, then simulate.
Example chain:

  1. "What's dragging my Google Ads account down this week?" → audit

  2. "Tell me more about the creative fatigue on the prospecting campaigns." → drill-in

  3. "If I pause the three most fatigued ads and shift that spend to the top-performing cohort, what does my forecast look like?" → simulation

That entire chain is minutes, not days.

05

Let uncertainty drive your conviction

MMM outputs return credible intervals, not point estimates. A tight interval is a confident recommendation. A wide interval tells you the data doesn't support a big bet yet - scale cautiously or run an incrementality test first.

06

Name the tool surface when you want a specific tool

Sometimes you already know which tool you want to hit. Call it by name in your prompt for faster routing.
Example: "Use the Meridian budget reallocation tool. Allocate an extra $27K/week weighted toward channels with headroom. Simulate 4 weeks."

07

Save your best prompts

Every marketing team develops a set of recurring questions: Monday morning diagnostic, mid-week budget check, Friday creative review. Save the prompts that work. Paste them as a team. The MCP's value compounds when the same question asked the same way returns the same structured answer every week.

Troubleshooting

"Connection failed" or "401 unauthenticated"

Your BlueAlpha auth token expired or was revoked. Re-authenticate from inside your MCP client - on first tool call after reconnection, you'll be prompted to sign in again. If that doesn't work, check that your BlueAlpha workspace is still active at app.bluealpha.ai.

"No MMM models available"

Your workspace doesn't have a fitted Meridian model yet. Either upload one via the BlueAlpha platform (meridian_upload_model) or ask your BlueAlpha contact to build one for you.

"Tool not found" after a BlueAlpha release

Restart your MCP host (Claude Desktop or Codex) so it re-indexes the tool surface. The hosted MCP doesn't require a re-install, but clients cache the tool list at startup.

Slow or timed-out responses on Google Ads

Large accounts can hit timeouts on wide queries. Scope your prompt to a date range (e.g., last 30 days instead of last year) or a specific campaign set.

Rate-limited

Starter tier has standard rate limits. Paid tiers get higher limits and priority routing. Enterprise is negotiated. See pricing or ask your account contact.

"Authentication loop" - Claude Desktop keeps re-prompting

Delete the cached auth in Claude Desktop's secure storage, restart the app, and reauthenticate. This is usually a stale token problem.

Plugin doesn't appear after install

Run claude plugin list to confirm install status. If it's there but not loading, run claude plugin reload bluealpha. If the plugin isn't in the list, the install didn't complete - reinstall and watch for error output.

Frequently asked questions

  • Which clients work?

    What data leaves our workspace?

    Do I need to know how to code?

    How is this different from the BlueAlpha platform?

    Can I simulate budget changes before committing spend?

    Does the plugin support Claude Desktop?

    How often do new capabilities ship?

    How often do new capabilities ship?