NeuronSearchLab MCP Is Live in the OpenAI Marketplace
NeuronSearchLab's MCP server is now available through OpenAI, giving teams a governed way to inspect recommendations, search catalogues, explain rankings, and operate discovery systems from ChatGPT.
NeuronSearchLab MCP Is Live in the OpenAI Marketplace
NeuronSearchLab's MCP server is now live in the OpenAI marketplace. That means teams can connect ChatGPT to their recommendation workspace and use governed tools for catalogue search, ranking inspection, contexts, rules, experiments, analytics, and operational debugging.
The important shift is not that a recommendation platform has another integration. It is that recommendation operations can now happen inside the same assistant environment where product managers, engineers, merchandisers, editors, and ML teams already ask questions about system behaviour.
For recommendation teams, that matters because the hard problems are rarely abstract. They are practical questions:
- What would this user see on this surface right now?
- Why did one item rank above another?
- Which catalogue records match this query?
- Which context, rule, or pipeline setting is shaping the result?
- Are experiment results moving because the model improved, the catalogue changed, or traffic shifted?
Those questions need access to live platform state, not only documentation or static reports.
Why MCP matters for recommendation systems
The Model Context Protocol gives AI clients a structured way to call external tools. For NeuronSearchLab, that means ChatGPT can interact with recommendation operations through a controlled server instead of relying on copied API keys, ad hoc scripts, or screenshots from a console.
A recommendation system is not only a model endpoint. It is a live operating layer made up of catalogue data, behavioural events, contexts, ranking pipelines, business rules, segments, experiments, training jobs, and analytics. Teams need to inspect and adjust those parts without turning every question into a backend ticket.
MCP is useful because it gives the assistant a defined tool boundary. It can search, inspect, explain, and, where permitted, make changes through explicit platform actions. The platform still owns authentication, permissions, audit boundaries, and the underlying API contract.
That makes it a better pattern than a generic chatbot bolted onto a dashboard. The assistant is not guessing from text. It is calling recommendation-specific tools.
What the NeuronSearchLab MCP server can do
The hosted NeuronSearchLab MCP endpoint is available at:
https://console.neuronsearchlab.com/api/mcp
It exposes a submission-safe operational profile for ChatGPT and other remote MCP clients. Depending on the authenticated workspace and scopes, teams can use it to:
- search catalogue items and inspect item details
- explain why an item ranked where it did for a user or context
- list and inspect recommendation contexts
- create and update ranking contexts
- review ranking pipelines and rule configuration
- inspect segments, experiments, training jobs, and analytics
- compare items and review top-performing catalogue records
- manage event types and integration visibility
- revoke API keys from the security surface when needed
The local npm package is still available for MCP clients that prefer stdio-based setup:
npx -y @neuronsearchlab/mcp
That gives teams two paths: a hosted endpoint for fast connection, or a local package for development environments and clients that still expect stdio.
Why the OpenAI listing matters
Being available through OpenAI changes the adoption path. Recommendation operations can now sit closer to the conversation where decisions are made.
A product lead can ask why a homepage feed changed. An engineer can inspect a context before touching code. A merchandiser can review whether a rule is affecting a launch surface. An ML operator can look at experiment results and training status without jumping through multiple tools.
This is especially useful for smaller teams that need serious recommendation control but do not have a large internal ML platform group. The goal is not to replace the console or the API. It is to make the first investigation faster and make the next action clearer.
The best version of agentic recommendation operations is narrow, auditable, and operationally useful. Read-only actions should be quick. Destructive or behaviour-changing actions should be explicit. Credentials should be scoped. Access should be revocable.
That is the model NeuronSearchLab is building around.
A practical demo path
A good first test is not a broad automation. It is a small operational loop that proves the assistant can work with real recommendation state.
Try this flow after connecting the MCP server:
Get 10 recommendations for user demo-user@example.com using context homepage-feed.
Search the catalogue for waterproof jackets and show the top 5 item ids.
Record a click event for one returned item using the request_id from the recommendation response.
Explain why that clicked item ranked where it did for demo-user@example.com.
That path checks four important capabilities: retrieval, catalogue search, feedback capture, and ranking explanation. It also verifies that attribution survives from the recommendation response into the event trail, which matters when teams later evaluate quality.
For a read-only first pass, skip the event recording step and focus on recommendations, search, ranking explanation, metrics, and context inspection.
How to connect
For ChatGPT, use the OpenAI marketplace listing for NeuronSearchLab. The hosted server uses OAuth so a user can sign in to the NeuronSearchLab console, approve access, and revoke connector-created keys later from the security page.
For other MCP clients, connect directly to the hosted endpoint:
https://console.neuronsearchlab.com/api/mcp
The full setup guide is in the NeuronSearchLab MCP documentation. The MCP package and hosted server details are also available in the NeuronSearchLab MCP repository.
Where this fits in the platform
NeuronSearchLab is building the operating layer for modern recommendation and discovery systems: catalogue ingestion, events, embeddings, retrieval, ranking, rules, contexts, experiments, analytics, and explainability.
The MCP integration brings that operating layer into AI clients. It gives teams a practical way to ask operational questions, inspect live ranking behaviour, and move from investigation to action without losing governance.
If your team is exploring AI agents for recommendation workflows, start with the MCP integration guide, then connect one real context and run the smoke-test loop above. For the broader platform surface, explore NeuronSearchLab features or the recommendation systems reading path.
FAQ
What is the NeuronSearchLab MCP server?
It is a Model Context Protocol server that lets MCP-compatible AI clients call NeuronSearchLab recommendation, catalogue, ranking, context, experiment, and analytics tools through a governed interface.
Does this replace the NeuronSearchLab console?
No. The console remains the main operational UI. MCP gives teams an assistant-driven path for investigation, inspection, and controlled actions from clients such as ChatGPT, Claude, Codex, Cursor, and Windsurf.
Can teams control what the assistant is allowed to do?
Yes. Access is governed by authentication, scopes, tool annotations, and revocable credentials. Read-only inspection can be kept separate from actions that change ranking behaviour or platform configuration.
Is a local install still available?
Yes. Teams can use the hosted endpoint at https://console.neuronsearchlab.com/api/mcp or run the npm package locally with npx -y @neuronsearchlab/mcp.