Your Wells, Inside Any Agent

Claude, Gemini, ChatGPT, or any MCP client calls hosted tools on your rod-lift wells, rank gearbox loads, diagnose wells, read dyno cards. Same credential and scopes as the REST API.

Read the MCP docs Request Demo

Claude

MCP What This Surface Is

Tools Your Agents Call on Your Wells

Model Context Protocol turns an AI client into a caller on your PetroBench org. The client asks in plain language. The server returns scoped tool results from the live wells.

A Hosted Tool Server

MCP is how agents discover and call tools. PetroBench hosts the server: Claude, Cursor, ChatGPT, or any MCP client connects to one URL.

Your Wells, Not a Demo Sandbox

Agents rank wells, diagnose loads, read dyno cards, and look up equipment on the same rod-lift records your engineers already use.

Same Token as the REST API

One scoped Bearer credential. Same org boundary and abilities as HTTP. No second identity store for agents.

Setup Connect a Client

Three Steps to a Live Tool Call

No local server to run. Mint a token, paste the hosted URL, and ask from Cursor, Claude, ChatGPT, Gemini, or any MCP client you already use.

Mint a scoped token

Create a personal or service token in Settings, or with the pb CLI. Pick the same scopes you would grant over REST.
Point your client at the hosted URL

Add https://app.petrobench.com/mcp in Claude, Cursor, ChatGPT, Gemini, or any MCP client that speaks streamable HTTP. Pass the Bearer token in the Authorization header.
Ask against your live wells

Rank hot wells, diagnose a leader, pull equipment. Tools run under your org scopes. Live-well writes stay behind engineer signoff.

mcp.json

# mcp.json, streamable HTTP

{
  "mcpServers": {
    "petrobench": {
      "url": "https://app.petrobench.com/mcp",
      "headers": {
        "Authorization": "Bearer pb_pat_..."
      }
    }
  }
}

# remote MCP connector

# Server URL
https://app.petrobench.com/mcp

# Authorization header
Bearer pb_pat_...

# Same scopes as REST, mint in Settings or pb CLI

# custom MCP / connector

# Server URL
https://app.petrobench.com/mcp

# Authorization header
Bearer pb_pat_...

# Wire the same URL ChatGPT uses for remote MCP tools

# remote MCP / extensions

# Server URL
https://app.petrobench.com/mcp

# Authorization header
Bearer pb_pat_...

# Same hosted endpoint Gemini MCP clients call

Exact client fields vary by host. Full wiring steps live in the MCP docs.

Capabilities On Your Wells

What Your Agents Can Do

Wells, simulations, equipment, and account reads, real PetroBench operations your agents call with the same scoped permissions.

Full MCP reference

Wells

Find the Right Wells

Search and rank the wells, pull production signal, and catch data gaps before a sim or diagnosis.

Search & rank
Production trends
Data gaps
Well readiness

Compute

Run Sims & Diagnose

Load dyno cards, diagnose wells, compare designs, and wait on long jobs, same engine as the workspace.

Dyno cards
Diagnostics
Design compare
Job polling

Catalogs

Look Up Equipment

Resolve pumping units and rods from the same catalogs your engineers design against.

Pumping units
Rod strings
Installed gear
Custom items

Account

Stay Inside Limits

Keep agents paced against org usage, contacts, and handoff context.

Usage & limits
Contacts
Account profile

More Beyond MCP

One Engine. Three Surfaces

Prefer raw HTTP, or chat with Vector on the same engine. Same credential and scopes.

Prefer raw HTTP

See the API.

Same wells, simulations, and equipment over HTTP, same credential and scopes as MCP.

Versioned REST

Explore the API

Want chat on the engine

Work With Vector.

Vector sits on the same rod-lift engine. Chat through sims, dyno cards, and drafts in your workspace.

In your workspace

Open Vector

Put Your Wells One Call Away

Enterprise includes MCP access. Professional adds it through the API access add-on. Connect Claude, Cursor, or your own agent in an afternoon.

Démo de simulation en direct

Vos données de puits

Équipe d'ingénierie

Des questions ou besoin d'aide pour autre chose ? Contactez-nous