Today we're opening the PetroBench API and MCP server to all customers. Both have been in early access with a group of major operators since earlier this year. They're now generally available.
The API and the MCP server
The API connects your systems to the platform. Wells, equipment, simulations and results are available over REST, and webhooks notify your systems when something changes. Data teams use it to keep Snowflake, Databricks or Azure Event Hub current without exports.
The MCP server connects your AI tools to the same platform. Connect Claude, ChatGPT, Gemini, Grok or any other MCP-compatible tool, and your agent can find wells, read and update equipment, run simulations and read results. The agent decides the sequence.
Both use the permissions you already manage. A token can't do more than the user who owns it, and every call is recorded in the audit log alongside actions from the browser. SSO, SCIM and MFA apply to agents the same way they apply to people.
Examples from the pilot
A few things pilot users did from their own AI tool:
- Ask which wells in a field haven't been simulated this year, and get the list with the last run date on each.
- Pull the installed rod string on a well, run it at a higher stroke rate, and read the loads back in the chat.
- Compare the last two designs on a well and summarize what changed and why the loads moved.
- Update the tubing anchor on a well after a workover, then rerun the installed design.
- Load every simulation result from the past month into the warehouse, every night, without an export.
Each of those runs under the user's own permissions and shows up in the audit log like any other action.
Getting started
Book a demo at petrobench.com/demo and we'll run it on your wells.
Already a customer: open Connected apps in Settings and connect your AI tool to app.petrobench.com/mcp. Docs are at docs.petrobench.com.
Where this is going
We started PetroBench more than five years ago on one idea: rod lift engineering belongs on a shared platform in the cloud, with the wells, the equipment and the people in one place. Today's release exists because of that decision. The API and the MCP server are the same platform, reached from another tool.
Because the platform already holds the data, runs the simulations and enforces the permissions, adding a new way in means exposing what's already there. That's what let us go from a pilot with major operators to general availability without building a separate product.
Our bet is that more production engineering work runs through systems and agents over the coming years, with engineers setting the standards and reviewing the results. The platform is built to carry that across an operator's full well count. Every capability we ship from here arrives for people, for systems and for agents at the same time, on the same foundation we've been building on since the start.
Related: How to Connect Claude to the PetroBench MCP Server. How Agent Permissions Work on PetroBench.