∇ProvenDecisions
Ctrl K

Optimization API and MCP server

Use ProvenDecisions from your own programs and AI assistants. The API takes a model as typed math or LaTeX and returns JSON: the solver's status, the objective, the solution, the check of every constraint, the dual or KKT certificate and the proof label, with the step-by-step walkthrough at the depth you choose. The MCP server gives the same tools to Claude and other assistants, so they quote a solver's checked numbers instead of guessing. Included in the Standard and Pro plans.

Get a key

Sign in, open Settings, and under API keys choose Create key. The key (it starts with pd_live_) is shown once: copy it then. We keep only a scrambled copy, so a lost key cannot be shown again; revoke it and make a new one. Keys are personal: keep them out of shared code and public places.

Solve a problem

The product mix example: make products A and B to earn the most profit ($) within two capacity limits.

curl -s https://proven-decisions.ai/api/v1/solve \
  -H "Authorization: Bearer $PD_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "maximize 3*a + 2*b\nsubject to\n2*a + 1*b <= 100\n1*a + 2*b <= 80\na, b >= 0", "depth": 3}'

The answer (shortened):

{
  "status": "optimal",
  "objective": 160.0,
  "solution": {
    "a": 40.0,
    "b": 20.0
  },
  "proof_label": "Exactly proven optimal",
  "verification": {
    "feasible": true,
    "max_violation": 0.0,
    "verified": true
  }
}

The same in Python:

import requests

KEY = "pd_live_..."   # Settings, API keys
model = """maximize 3*a + 2*b
subject to
2*a + 1*b <= 100
1*a + 2*b <= 80
a, b >= 0"""
r = requests.post("https://proven-decisions.ai/api/v1/solve", json={"model": model, "depth": 3},
                  headers={"Authorization": f"Bearer {KEY}"}, timeout=60)
r.raise_for_status()
ans = r.json()
print(ans["status"], ans["objective"], ans["solution"], ans["proof_label"])

Endpoints

Errors come back as JSON {"detail", "code"}: 400 (not typed math, empty), 401 (missing or revoked key), 403 (plan without API access), 413 (too long), 422 (the model could not be read) and 429 (a limit is reached; the message says which and when it resets).

Add the MCP server to Claude

Claude Code (terminal or VS Code), signing in with no key: add the server, then type /mcp in a new session, choose it and choose Authenticate; your browser opens our sign-in page, where you choose Allow.

claude mcp add --transport http --scope user proven-decisions https://proven-decisions.ai/mcp

Or with a key (for example on a server with no browser):

claude mcp add --transport http --scope user proven-decisions https://proven-decisions.ai/mcp \
  --header "Authorization: Bearer $PD_KEY"

Claude Desktop: add this to claude_desktop_config.json (Settings, Developer, Edit config; it uses the mcp-remote bridge, so Node.js is needed), then restart Claude:

{
  "mcpServers": {
    "provendecisions": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://proven-decisions.ai/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer pd_live_..."
      }
    }
  }
}

Tools: solve, dual, kkt and explain (the walkthrough as text at a depth you choose), and for the site's content search_content, get_content and list_content. Other assistants that speak MCP over streamable HTTP and can send an Authorization header connect the same way, at https://proven-decisions.ai/mcp.

Connect from claude.ai

On claude.ai and in the Claude apps, add ProvenDecisions as a custom connector; you sign in instead of copying a key:

  1. In Claude, open Settings, then Connectors, and choose Add custom connector.
  2. Name it ProvenDecisions and enter the address https://proven-decisions.ai/mcp. Keep the choices Claude marks Detected: Sign in now and Register automatically; add no request headers.
  3. Choose Connect. A ProvenDecisions page opens: sign in with your ProvenDecisions account if asked.
  4. It asks whether to allow Claude to use ProvenDecisions on your account. Choose Allow; you go back to Claude with the connector on.

The connector gets the same tools and the same plan rules as a key: Standard and Pro, your plan's limits shared with the website, typed math only for solving. Claude stays signed in (access renews by itself for up to 30 days of no use). To stop it, choose Disconnect under Settings, Connected apps in ProvenDecisions, or remove the connector in Claude. On a Team or Enterprise plan, an owner adds the connector for the organization first.

Limits

API and MCP requests share your plan's limits with the website; a script cannot do more than you can in the app: at most 5 requests a minute, 100 a day and 1,000 a month; one problem at a time; 10 minutes of solving a day; one problem per request, up to 3,000 characters; and the monthly stop at 85% of your plan's price. Typed math and LaTeX only for now: word, photo and voice problems are solved in the web app. Vector syntax (var x[100]) is part of Pro, as on the web.

More: API reference (every endpoint, error and MCP tool) · OpenAPI document · API and MCP in Help · Plans and limits · Pricing · Terms

Try ProvenDecisions free: type, photograph or say an optimization problem; every answer is checked by a solver.

What do you want to optimize?

Type the math, describe it in words, snap a photo or say it. ProvenDecisions builds the model, solves it, checks it and walks you through it.

One constraint per line, or separated by commas. Add integer: x, y or binary: z for whole-number decisions. LaTeX works too.

Try: