Mineralis

2. Use Mineralis from an AI agent (MCP)

The Mineralis MCP server puts this API in front of an agent: it can look up the exact mineral, element and environment IDs, check how a CSV's columns map and what a run would cost, run a free preview of up to 1,000 samples, and — only when you confirm the cost — run a charged batch against this account. It runs on your own machine, talks to this service over HTTPS, and does nothing you could not do with curl.

Assay data is passed to it by file path and read locally, so a drillhole table is never pasted into a conversation. Results come back summarised; the complete table is in the CSV this service emails you.

Install it with one command

Node 20 or newer is required. The script writes one folder (~/.mineralis-mcp), runs npm install in it, and registers the server with Claude Code if the claude command is on your PATH.

Or do it by hand: fetch the files listed in the manifest into a folder, run npm install there, and register node <folder>/src/server.mjs using one of the blocks below.

claude mcp add mineralis --env MINERALIS_API_TOKEN=YOUR_TOKEN -- node /path/to/mineralis-mcp/src/server.mjs

{
  "mcpServers": {
    "mineralis": {
      "command": "node",
      "args": ["/path/to/mineralis-mcp/src/server.mjs"],
      "env": { "MINERALIS_API_TOKEN": "YOUR_TOKEN" }
    }
  }
}

[mcp_servers.mineralis]
command = "node"
args = ["/path/to/mineralis-mcp/src/server.mjs"]

[mcp_servers.mineralis.env]
MINERALIS_API_TOKEN = "YOUR_TOKEN"

[mcp_servers.mineralis.tools.mineralis_run]
approval_mode = "approve"

The install includes README.md and AGENT_GUIDE.md: the first covers the environment variables, the second is written for the agent and documents every tool parameter. The bridge also serves the guide to the agent as an MCP resource, so it can read it without leaving the session.

3. POST to the API yourself

POST an execution packet as YAML or JSON to run Mineralis from a script. Runs are billed to this account at the same rate as runs started from this page.

For a large batch, send the assay data as CSV instead of inline YAML: multipart/form-data with a parameters part (the settings only) and a data part (the CSV, first row the column IDs). The same packet is three to four times smaller that way, gzip is understood on either part, and the assay values are read straight from the file rather than parsed as a YAML document. Every rule is the same for both forms — one Sample ID column, one column per property.

A result too large to return in one response comes back as its first rows plus a link to the complete table as a CSV file, which is also emailed to you. The link is valid for 48 hours.

Sign in to create an API token.

Every request is a separate run and a separate charge. If a request times out, check Run history on the User Administration screen before sending it again — the run may have completed and been billed.