The Optilogic MCP Connector

Overview

The Optilogic Model Context Protocol (MCP) Connector links AI Agents like Claude and ChatGPT to Ada, Optilogic’s agentic AI for supply chain modeling. Together, they give teams a faster way to make better tactical decisions within today’s supply chain — and design the supply chain they need for tomorrow. Ada works with a live digital twin of your supply chain, combining mathematical optimization, simulation, and demand modeling to answer questions across the full planning horizon, from day-to-day operational response to long-term network, transportation, inventory, and production strategy.

Once connected, the AI Agent can list the model databases in your Optilogic account, open a conversation with Ada, attach one or more databases, and relay prompts and responses back and forth — all from inside the agent. This enables AI-powered what-if analysis, demand and sourcing analysis, routing and inventory tradeoffs, tariff scenarios, and network strategy without switching tools.

In practice, Ada continues to do what she does best — reasoning over your supply chain data, running analyses, and answering modeling questions. The AI Agent adds a complementary layer on top: turning Ada’s outputs into decision-ready executive summaries, spreadsheets, slide decks, and interactive dashboards, while combining them with web research and other connected tools in a single workflow.

The connector is currently available in Claude (Anthropic), ChatGPT (OpenAI), Grok (SpaceXAI), and Vibe (Mistral AI).

‍

Quick Start

1. Add the Optilogic connector to the AI Agent you are using and authenticate with your Optilogic account. When prompted for the MCP Connector URL, enter: https://mcp.optilogic.app/mcp. Direct links to set up a (custom) connector for the AI Agents that currently support the Optilogic MCP Connector are:
   1. Claude: https://claude.ai/directory/optilogic
   2. ChatGPT: https://chatgpt.com/plugins (note that Developer Mode needs to be turned on prior to connecting)
   3. Grok: https://grok.com/connectors
   4. Vibe: https://chat.mistral.ai/connections
2. In a chat with your AI Agent, ask it to list the databases in your Optilogic account to confirm the connection.
3. Tell the agent which database(s) (Cosmic Frog models, DataStar projects, or other Postgres databases) you want to work with, by name. If you are unsure, just state your question and the agent can help you identify which database(s) will help answer.
4. Ask your question or describe your task in plain language. The agent will start an Ada session, attach the right database(s), and relay your prompt.
5. Review Ada's answer, ask follow-up questions, and — when you are ready — ask to turn the findings into a document, spreadsheet, deck, or living dashboard.
6. Revisit or continue conversations anytime, plus you can see (and continue) these interactions with Ada and any work done by Ada on the Optilogic platform.

Detailed step-by-step instructions for Claude and ChatGPT, including screenshots to (dis)connect, can be found here:

• Getting Started with the Optilogic MCP Connector in Claude
• Getting Started with the Optilogic MCP Connector in ChatGPT

Understanding the Optilogic MCP Connector

What Is It?

The Optilogic MCP Connector is a Custom Connector for AI Agents built on the Model Context Protocol (MCP). It gives AI Agents a set of tools that let it act as an orchestrator for Ada conversations: discovering your models, starting and managing Ada sessions, attaching databases, and polling for Ada's (asynchronous) responses.

It is not a replacement for Ada or for the Optilogic platform — it is a bridge. Ada still does the actual modeling work; the connector simply gives the agent a way to ask it questions and receive answers.

A useful mental model: the AI Agent is the orchestrator and communicator; Ada is the subject-matter expert on your models.

What Should I Use It For?

Teams are using the Optilogic MCP Connector for tasks like:

• Interrogating scenario outputs — cost drivers, service-level trade-offs, which facilities open or close
• Comparing multiple models or scenarios side by side
• Sanity-checking model data for missing values, outliers, or duplicates before a solve
• Turning Ada's analysis into executive summaries, board-ready briefs, and slide decks
• Building scenario-comparison workbooks and interactive dashboards from model outputs
• Filling gaps in a model with outside context — for example, pulling real-world wage or tariff data to round out a scenario

The Connector works best for grounded, model-based questions where Ada supplies the underlying facts and the agent handles synthesis, formatting, and communication. It is less suited to open-ended business strategy discussions that have no connection to an actual model or dataset - “What happens to warehouse utilization if demand is up 5% across category A?” is only meaningful when asked in the context of a model.

What the Optilogic MCP Connector Can and Cannot Do

*To learn more about using Optilogic Teams, please see this Getting Started with Optilogic Teams Help Center article.

**To learn more about Ada’s interaction style and agent options, please see the Create Your First Prompt section in the Getting Started with Ada & Agentic AI Help Center article.

Prior to Setting Up

Before you set up and start working with the Optilogic MCP connector, please take note of following:

• To use the connector, you need to have an Optilogic user account and a user account on the AI Agent you want to use the connector in.  
  • You can create an Optilogic account on the Create a Free Account page (step-by-step instructions are here)
• The AI Agents are third-party services governed by your terms with the AI Agent’s company.
  • See for example Anthropic’s consumer and commercial terms of use.

Tools and Descriptions

The following tools are available to the connector:

Agent Lifecycle

Past Conversations

Artifacts

Workspace Files

‍

Sharing

Account / Teams

Data

‍

Example Prompts by Use Case

The prompts below are starting points — swap in your own model names, regions, priorities, etc.

Getting Oriented

• Inventory your models: “Using the Ada connector, list all the models in my Optilogic account and group them by what they appear to be for (production models, tests, training exercises).”
• Profile a model before diving in: “Start an Ada session with my ‘[MODEL NAME]’ model attached. Ask her to summarize the model: what tables it has, how many customers/facilities/products, what scenarios exist, and whether there are solved outputs.”

Analysis & Insights

• Interrogate scenario outputs: “Ask Ada (with ‘[MODEL NAME]’ attached): which scenario has the lowest total cost, what drives the difference vs. baseline, and which facilities open or close in each scenario? Then summarize her answer as a comparison table.”
• Compare across models: “Start an Ada session with both ‘[MODEL A]’ and ‘[MODEL B]’ attached. Ask her how the two models differ in network structure, demand, and assumptions, and which is more current.”
• Sanity-check model data: “Ask Ada to profile the data in ‘[MODEL NAME]’: missing values, orphaned records, duplicate names, suspicious outliers in demand or costs. Have her rank issues by severity, then give me a cleanup checklist.”

Deliverables — Where AI Agents Shine

• Executive summary from model outputs: “Ask Ada about the scenario results in ‘[MODEL NAME]’ (costs, service levels, key network changes). Then write a 1-page executive summary as a Word doc for leadership — plain language, decision-focused.”
• Scenario comparison workbook: “Get scenario-by-scenario cost and flow summaries from Ada for ‘[MODEL NAME]’, then build an Excel workbook: one tab per scenario, plus a comparison tab with deltas vs. baseline and a chart.”
• Results readout deck: “Interview Ada about ‘[MODEL NAME]’ — objectives, scenarios tested, key results, recommendation. Then build a 5-slide PowerPoint readout: context, approach, results, tradeoffs, recommendation.”

Broader Supply-Chain Scenarios to Try

• Network redesign trade-offs — build several distribution network scenarios and summarize cost, service-level, and risk trade-offs for the board
• Cost-to-serve diagnosis — run a cost-to-serve analysis by segment/channel and flag unprofitable segments
• Disruption stress test — simulate a supplier or port disruption and write an executive risk memo
• Greenfield site selection — weigh candidate sites against strategic priorities and draft a siting recommendation
• Reshoring / nearshoring business case — model a production shift and produce a cost-benefit summary
• Tariff shock scenario — re-run sourcing with a tariff increase and quantify the cost impact
• Demand shift scenario — model a demand shift and recommend how to rebalance the network

‍

Best Practices

• Name the Model Explicitly: Ada only sees databases attached to the conversation. Telling the agent the exact model name(s) ensures the right one(s) get attached the first time. If you are unsure of which database to use, ask the agent and you can pick from a list.
• Chain, Do Not Cram: Ask Ada focused questions in sequence rather than one giant prompt. Because sessions keep context, follow-up questions are cheap — you do not need to restate everything each time.
• Let Each Side Do Its Job: Use Ada for anything touching model data, and the agent for formatting, synthesis, web research, and file creation. Trying to get the agent to reason about model internals directly, without Ada, will produce weaker answers.
• Capture the Raw and the Polished: Before you ask the agent to build a polished deliverable, ask it to show you Ada's raw answer first. This makes it easy to catch a misreading early, before it propagates into a document or deck.
• Multi-Attach for Comparisons: When you need cross-model answers, one Ada session with two databases attached beats juggling two separate sessions.
• Give Context Before the Task: As with Ada directly, better prompts include business context, constraints, and the specific task — not just the task alone.
• Start Fresh for a New Topic: If you are switching to an unrelated model or question, start a new conversation rather than overloading one thread.
• Ask for Multiple Options: Instead of one recommendation, ask Ada (via the agent) for several approaches and their trade-offs.
• Supplement Missing Data: If Ada flags gaps in the model, ask the agent to search the web for relevant proxy data to round out the story — and to cite where it came from.

‍

Data Handling and Privacy

When you use the Optilogic MCP Connector, prompts and the data Ada returns pass through the agent in order to be displayed, summarized, or turned into a deliverable. This is in addition to — not a replacement for — Optilogic's own data handling for Ada itself.

• Data users share with the AI Agent is governed by the AI Agent company’s privacy policy; Optilogic's privacy policy applies once data reaches Ada on the Optilogic platform.
  • AI Agent privacy policy example: Anthropic (Claude) Privacy Policy
  • Optilogic Privacy Policy
• Optilogic’s data policies can be found here: Optilogic Trust Center; Optilogic’s specific data policies for AI can be found here: AI Data Security and Privacy.
• As with any connector, only attach the database(s) you intend to discuss in a given conversation.
• Avoid including sensitive information (PII, credentials, etc.) in prompts, table names, or column names, since these may be passed to the underlying AI systems on both sides of the connection.

What the MCP Server Collects and How it is Used

The MCP Server connects AI agents to Optilogic's Public APIs and Ada, Optilogic’s native AI agent. Ada and the Public APIs have logging in place; MCP server calls are logged in accordance with those practices with additional meta data to identify it as coming from the MCP, nothing additional - it does not have any user identifiers in it.

The MCP Server only processes the tool-call payloads needed to fulfill the requests; surrounding conversation context from the agent is not retained. In addition to serving the request, tool-call data is also used for the following purposes:

• Debugging: MCP request/response data and associated logs may be reviewed to investigate errors and user-reported quality issues.
• Analytics: Optilogic collects usage data (such as prompt counts, session activity, token consumption, timestamps) associated with the user account to understand adoption and improve the service.

For clarity, Optilogic does not use logs for model training.

Data Retention

MCP request/response data and associated logs are stored in accordance with Optilogic's Data Retention Policy. You can learn more here -- How We Safeguard Your Data: Backups & Retention Explained.

What happens when Disconnecting the Connector

Disconnecting the Connector on the AI Agent’s settings tells the agent to stop using and refreshing your Optilogic tokens — that is the immediate effect.

After disconnecting, the historical chats that were had through the connector are still accessible to the user, just no follow-up prompts that require accessing Ada can be added.

Visibility on the Optilogic Platform

Please note that users have visibility into the MCP Connector conversations and actions on the Optilogic platform in following places:

• Users can view the calls made by the MCP Server to Ada in their Optilogic account Chat history; they are tagged as MCP.
• Users can view model run activity from the MCP in their Optilogic account as a part of their broader account usage under the Account > Usage screen.

Troubleshooting

If the conversation with Ada through the AI Agent seems to go off track, e.g., gives no response or odd/incorrect responses, please follow these troubleshooting steps:

1. Ensure you are logged into the correct Optilogic account (in case you have multiple Optilogic accounts) and are working in the correct workspace (My Account vs a Team account). You can test this by asking the agent which Optilogic account and which workspace it is logged into. If logged into the wrong Optilogic account, go to bullet #5; if working within the wrong workspace, ask the agent to switch to the correct workspace. If not resolved, then:
2. Double-check the data you are trying to access exists in the logged in account and workspace on the Optilogic platform itself. If not resolved, then:
3. Start a new conversation – this is always recommended when switching between Optilogic accounts / workspaces / databases / major tasks. If not resolved, then:
4. Do a hard refresh of your browser (Ctrl +Shift + R on Windows, Cmd + Shift + R on Mac) / restart your desktop application – this may be needed when a newer version of the connector becomes available. If not resolved, then:
5. Disconnect and reconnect the Optilogic MCP Connector.
   1. Note that this is required when switching between Optilogic accounts.
6. If you are still having trouble, please contact the Optilogic support team on support@optilogic.com.

Example Conversation Walk-through

Example Overview

In this example, we will:

1. Connect to a Hopper model named Territory Planning
   1. This model can be copied to your account from the Resource Library: Territory Planning model on the Resource Library
   2. The model is documented in this Territory Planning (Transportation Optimization) Help Center article
   3. We made 1 change: removed the Distance Cost of $1/MI in the Transportation Rates table
2. Inventory what is in the model in terms of inputs and scenarios
3. Identify any missing data and potentially useful scenarios to add
4. Fill the data gap using publicly available sources
5. Add several scenarios based on the recommendation
6. Run all scenarios
7. Generate a summary of the outputs, including an interactive map showing the routes by scenario

Claude was used in this example; using the same prompts in another AI Agent will result in similar responses.

Summary of Prompts and Responses

The prompts used and a summary of the responses follows here, see the appendix for the full conversation captured in screenshots. Note that the some of the “Please check” prompts to check back in with Claude if there is a response from Ada yet when Claude has stopped polling are omitted here.

Prompt 1: Using the Optilogic Ada connector, can you review the inputs of the Territory Planning model? Please give me a summary and also check which costs are being modeled.

Response:

• High-level description of the type of model and its scenarios
• Summary of the model inputs, including row counts and notes on the data in the tables
• Overview of costs populated and those available but not populated
  • The only cost used is fixed cost per route, and no variable transportation cost is included
• Summary of findings

Prompt 2: Can you find what the average per mile cost for trucks in the Atlanta region is and add this to the model?

Response:

• Found national spot dry van rates at about $2.30-2.70/mile
• Atlanta region runs above national average at about $3.20/mile
• Flags that these are long-haul, full-truckload linehaul rates, whereas the model is a local delivery/routing model where cost drivers are different and a per mile cost is harder to find
• Suggests 2 reasonable choices for Distance Cost. The user chooses $3.20/mile, the Southeast regional spot freight rate.

Prompt 3: Before re-running, can you suggest a top 3 of additional scenarios that would be interesting to run, based on the expected value / insights they may provide?

Response: lists 3 sets of suggested scenarios to add and the reasoning:

• Time cost + stop cost added, re-run across territory counts
• Territory count sensitivity – sweep beyond 3 and 5
• Truck capacity/fleet size stress test

Prompt 4: Please add the scenarios for your #2 suggestion, territory count sensitivity, then run all scenarios (Hopper). Once done running, please create the trade-off curve for number of territories vs cost and an interactive map where the multi-stop routes of each scenario can be visualized, including tooltips and main KPIs by scenario.

Response:

• After some time working on the request, the user is asked to confirm the logic for the minimum delivery locations per territory for the new scenarios and the scenario naming. The user approves.
• The result is a downloadable interactive html file which contains Route Map and Cost Curve tabs. Users can switch between the scenarios and hover over the map / chart to bring up a tooltip with information about the route / scenario.

Other Helpful Resources

• Getting Started with the Optilogic MCP Connector in Claude
• Getting Started with the Optilogic MCP Connector in ChatGPT
• Getting Started with Ada & Agentic AI
• Modeler AI Agent

Questions or feedback on the connector? Reach out to the Optilogic Support team on support@optilogic.com. In addition, you can use the thumbs-up and thumbs-down buttons in the AI Agent chat to send feedback directly to the AI Agent’s company on any specific response.

Appendix – Example Conversation Screenshots