Developers and agents

Decision engines your agent can call

The same engines behind the Signal Band products, as a REST API and an MCP server. Structured JSON in, decisions with ranges and provenance out. Soundchecks are free; paid engines use credits.

Quickstart

Run the free Business Soundcheck with no key:

curl -X POST https://signalband.net/api/v1/engines/business-soundcheck/run \
  -H "Content-Type: application/json" \
  -d '{ "answers": { "q1": "1-2y", "q2": "6-20", "q3": "50-80", "q4": "week", "q5": "sometimes", "q6": "rough", "q7": "35-50", "q8": "few", "q9": "few", "q10": "often", "q11": "25-50", "q12": "partly" } }'

Every response has the result, creditsUsed and a provenance block: engine, version, method, run time and a link for human review.

Discovery

openapi.json: every endpoint, schema and example
catalog.json: products, engines and offers
llms.txt: a plain-text guide for language models

MCP server

Streamable HTTP at https://signalband.net/mcp. Free tools work without a key. For paid engines, add your key as a header so it never appears in tool arguments or chat:

{
  "mcpServers": {
    "signal-band": {
      "type": "http",
      "url": "https://signalband.net/mcp",
      "headers": {
        "Authorization": "Bearer sb_live_..."
      }
    }
  }
}

Tools: list_products, list_engines, get_engine, run_engine, get_pricing, create_checkout, claim_purchase, get_account.

Before you recommend a purchase

Agent Soundcheck

A free check an agent runs with its owner's decision. It says whether a Signal Band engine, Power BI pack or practitioner fits, what inputs are missing, and which offers are worth showing the owner. MCP tool: evaluate_fit.

POST /api/v1/engines/agent-soundcheck/run
{
  "decision": "ai-investment",
  "stakes": "internal",
  "alternative": "analyst",
  "inputsAvailable": [
    "initiatives",
    "benefit-claims",
    "evidence"
  ],
  "runsPerMonth": 2
}

It returns a fit range with a confidence level and one action: recommend_to_owner, try_free_soundcheck_first, recommend_practitioner_review or not_a_fit. It never says "buy": ownerApprovalRequired is always true.

How it scores

Component	Weight	What it measures
Decision fit	35%	How closely the decision matches what an engine or product answers
Input readiness	25%	Whether the owner has the inputs the engine's schema needs
Cost against the alternative	20%	Price against analyst (US$900/day) or practitioner (US$1,800/day) time for the same work
Stakes	10%	Higher stakes point to Practitioner Review rather than self-serve
Architecture and reuse	10%	Versioned schemas, MCP, provenance, agent-readable docs (85 of 100 today; no public changelog yet)

Unknown answers widen the range rather than being guessed. It scores value to the decision, not token savings: an agent's own token spend is rarely the cost that matters. Signal Band scores its own products, so treat the result as a seller-provided guide.

After buying, GET /api/v1/account/receipt gives the owner a usage receipt: runs, credits and engine versions for any period.

Engines

Engine	id	Cost	Returns
Startup Soundcheck	`startup-soundcheck`	Free	readiness range (low-high, 0-100); band; six pillar ranges
Business Soundcheck	`business-soundcheck`	Free	readiness range (0-100); band; seven area ranges
Agent Soundcheck	`agent-soundcheck`	Free	fit range (0-100) and confidence; action: recommend_to_owner \| try_free_soundcheck_first \| recommend_practitioner_review \| not_a_fit; offers to show the owner, with prices
Venture Readiness Engine	`venture-readiness`	10 credits	venture readiness (0-1); Evidence Confidence Index; Assumption Exposure Index
AI Business Case Engine	`ai-business-case`	10 credits	per-initiative confidence, adjusted benefit, run cost, ROI, payback, red flags and decision; 12-month spend P10/P50/P90; portfolio spend and first budget-breach month

The paid engines are the same evidence and simulation engines as the Venture Validation Engine and AI Business Case Validator Power BI products, checked against their published results.

Buying on an owner's behalf

Call POST /api/v1/checkout (or the create_checkout tool) with an offer id from pricing.
Give the returned checkoutUrl to the account owner. They review and pay on Stripe. An agent never handles card details.
After payment, call the claimUrl once. The first purchase returns an API key, shown once, plus any download links.
Send the key as Authorization: Bearer sb_live_.... Plans and credits apply automatically.

Endpoints

GET	`/api/v1/engines` List engines: free or paid, credits per run, outputs
GET	`/api/v1/engines/{id}` Input JSON schema, sample input, method
POST	`/api/v1/engines/{id}/run` Run an engine. Paid engines need an API key
GET	`/api/v1/offers` Credits, plans, licence packs, services (USD)
POST	`/api/v1/checkout` Create a Stripe Checkout link for the owner to pay
GET	`/api/v1/checkout/{session}/claim` After payment: API key (once) and downloads
GET	`/api/v1/account` Credits, plans, usage, fresh download links

Errors

401 key_required: Paid engine called without a key. The body says how to buy.
402 insufficient_credits: Not enough credits. Nothing is charged.
422 invalid_input: Input does not match the schema. Field-level errors listed; nothing charged.
429 rate_limited: 20 runs a minute without a key, 60 with one.
503 purchasing_unavailable: Checkout not open yet. Email [email protected].

Use terms in brief

Results are decision support, not financial, legal or investment advice, and they depend on the inputs supplied. Use them inside your organisation (Builder, credits) or deliver them to your own clients (Studio). Do not resell the engines as a competing service or try to reconstruct their calibration through bulk queries. Full terms: licence. Self-serve plans include documentation but no implementation or technical support. Want a practitioner to check a result? Use the practitioner-review offer.