▸ Developer Hub · steadywrk.dev

Build on SteadyWrk.

An HTTP API and an MCP server for AI field-service dispatch. The same 6 primitives — quote, estimate, order, evals, index, readiness_probe — callable from a web app, a script, Claude Desktop, or any MCP client.

OpenAPI 3.1 spec, no client library required. Sub-2hr quote turnaround, 90% completion rate. Machine-readable evals are a first-class citizen at every integration layer.

Quote turnaround: <2hr target
Completion rate: 90%
Dispatch latency p50: 340ms
Human override rate: 3%

Canonical estimates only (estimated:true). Recompute at /evals. Not a third-party audit.

▸ Install

Add the hosted MCP server.

Streamable HTTP at https://steadywrk.app/api/mcp. Keyless first: agentic.readiness_probe, dispatch.estimate, dispatch.evals, dispatch.index. Quote and order use POST /api/mcp/oauth/register or a key from hello@steadywrk.app. Registry publication of an npm package is a separate human step — these cards wire the live host.

Claude

Settings → Connectors → add https://steadywrk.app/api/mcp. Keyless: probe, dispatch.evals, index, estimate. Quote/order: POST https://steadywrk.app/api/mcp/oauth/register then POST https://steadywrk.app/api/mcp/oauth/token, or hello@steadywrk.app.

Open Claude connectors

Claude Desktop mcpServers JSON

{
  "mcpServers": {
    "steadywrk-dispatch": {
      "type": "http",
      "url": "https://steadywrk.app/api/mcp"
    }
  }
}

Cursor

Add to Cursor opens cursor://anysphere.cursor-deeplink/mcp/install (Cursor desktop, not a web page). name=steadywrk-dispatch is the mcp.json key — not registry id app.steadywrk/mcp-dispatch. If the link does nothing, paste the JSON into ~/.cursor/mcp.json or %USERPROFILE%\.cursor\mcp.json. dispatch.evals is public.

Add to Cursor

Cursor mcp.json snippet

{
  "mcpServers": {
    "steadywrk-dispatch": {
      "url": "https://steadywrk.app/api/mcp"
    }
  }
}

ChatGPT

Developer-mode Connectors: paste https://steadywrk.app/api/mcp. No Apps Store listing is claimed. Keyless probe, dispatch.evals, index, and estimate run without a key.

Open ChatGPT

MCP Streamable-HTTP URL

https://steadywrk.app/api/mcp

▸ Agent discovery

How agents find this API.

Eight public paths, a scored sample at Index Edition 3, and keyless JSON at /api/public/agent-discovery. Completeness is not quality. Dispatch figures stay on /evals as estimated:true.

Agent discovery hub Index Edition 3 Operate it yourself

▸ Integration surfaces

Three integration surfaces.

Same dispatch engine underneath. Pick the surface that fits your stack.

REST APIGA

OpenAPI 3.1

Full OpenAPI 3.1 spec. The public dispatch surface — evals, trade index, quote, and order — is documented, versioned, and testable from the spec playground. JSON over HTTPS, x-api-key auth on writes.

—/api/dispatch/v1/quote — matched contractor + cost estimate
—/api/dispatch/v1/order — create + track work orders
—/api/dispatch/analytics/evals — rolling eval metrics

POST /api/dispatch/v1/quote HTTP/1.1
Host: steadywrk.app
x-api-key: swrk_live_...
Content-Type: application/json

{
  "trade": "hvac",
  "location": { "state": "CO", "zip": "80302" },
  "urgency": "routine",
  "nte_cents": 85000
}

MCP ToolsClaude Desktop ready

Model Context Protocol

Four dispatch primitives as native MCP tools. Point Claude Desktop, Cursor, or any MCP client at the live Streamable-HTTP endpoint and your agent can quote, order, and read evals in one tool call.

—dispatch.quote — matched contractor + cost estimate + ETA tier
—dispatch.order — create work order, get SW-{id} handle
—dispatch.evals — rolling 30d operational metrics
—dispatch.index — trade-rate catalog (public, no key)

// claude_desktop_config.json
{
  "mcpServers": {
    "steadywrk-dispatch": {
      "type": "http",
      "url": "https://steadywrk.app/api/mcp"
    }
  }
}

ClientsSpec-driven

Generate your own

STEADYWRK publishes no package on npm or PyPI. The OpenAPI 3.1 spec is the client surface: generate a typed client in any language, or call the REST endpoints directly. An MCP client needs neither.

—Not yet on npm or PyPI — request access: hello@steadywrk.app
—OpenAPI 3.1 spec available now at /openapi.yaml
—TypeScript + Python clients planned

# No STEADYWRK package exists on npm or PyPI.
# Generate a typed client from the spec instead:
#
#   npx @openapitools/openapi-generator-cli generate #     -i https://steadywrk.app/openapi.yaml -g typescript-fetch
#
# Available today:
#   OpenAPI 3.1 spec → /openapi.yaml
#   (generate a typed client in any language)
#   MCP endpoint     → /api/mcp

▸ MCP reference

MCP tool reference.

Each tool follows the MCP JSON-Schema spec. dispatch.evals and dispatch.index are public read-only — no key required.

Tool	Args	Returns	Auth
dispatch.quote	{ trade, location{ state, city?, zip? }, urgency?, nte_cents? }	Matched contractor · estimated cost (cents) · ETA tier · quote validity	x-api-key
dispatch.estimate	{ trade, location{ state, city?, zip? }, urgency? }	Keyless billed-cost estimate — wage_basis + billed_estimate breakdown, not a matched quote	Public
dispatch.order	{ trade, location{ state, … }, description, urgency?, nte_cents?, reference_id? }	Order ID · masked contractor reference · status URL	x-api-key
dispatch.evals	{ period?: "rolling_30d" }	Goal fulfillment · ops metrics · trust signals	Public
dispatch.index	{}	8 trade verticals · BLS wage + billed routine rate	Public
agentic.readiness_probe	{ url }	Storefront readiness snapshot · not an attested Store Guard grade	Public

▸ Quickstart

Full dispatch flow.

Quote, confirm, dispatch — three round-trips, over MCP JSON-RPC or REST. Both transports are the whole integration surface; there is no client library to wait for.

MCP · JSON-RPC/api/mcp

// 1. Get a quote — tools/call dispatch.quote
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "dispatch.quote",
    "arguments": {
      "trade": "plumbing",
      "location": { "state": "AZ", "zip": "85001" },
      "urgency": "routine",
      "nte_cents": 120000
    }
  }
}
// → matched contractor, estimated cost, ETA tier

// 2. Create the work order — tools/call dispatch.order
//    { trade, location, description, reference_id? }
// → order id, masked contractor ref, status URL

// 3. Read live evals — tools/call dispatch.evals
//    { "period": "rolling_30d" } — public, no key

// No SDK required — any MCP client speaks this directly.
// Questions: hello@steadywrk.app

REST · curlOpenAPI 3.1

# 1. Quote
curl -X POST https://steadywrk.app/api/dispatch/v1/quote \
  -H "x-api-key: swrk_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "trade": "plumbing",
    "location": { "state": "AZ", "zip": "85001" },
    "urgency": "routine",
    "nte_cents": 120000
  }'

# → matched contractor, estimated cost (cents),
#   ETA tier, valid_for_minutes

# 2. Order
curl -X POST https://steadywrk.app/api/dispatch/v1/order \
  -H "x-api-key: swrk_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "trade": "plumbing",
    "location": { "state": "AZ", "zip": "85001" },
    "description": "Leaking supply line, suite 210"
  }'

▸ Authentication

Signed header for write calls.

All write operations require a live API key. dispatch.evals is public read-only. Keys are scoped per-account and revocable at any time.

Header format

x-api-key: swrk_live_<token>

MCP endpoint (no client key — public tools open, keyed tools proxied server-side)

"steadywrk-dispatch": { "type": "http", "url": "https://steadywrk.app/api/mcp" }

—Keys are prefixed swrk_live_ in production, swrk_test_ in sandbox.
—Rate limit: 100 req/min per key. Burst to 200 for 10s.
—dispatch.evals accepts no key — public endpoint, no rate limit.
—All decisions are logged. Append-only audit trail (7-year retention) queryable via API.

▸ For AI agents

A2A skill manifest.

The agent card at /.well-known/agent-card.json exposes four skills. dispatch.evals and dispatch.index are public read-only — no key, no auth.

dispatch.evalsPublic

Return the public operational evals for the rolling window: completion rate, NTE accuracy, dispatch latency (p50/p95), human-override rate. Self-reported, estimated from production telemetry.

{"skill":"dispatch.evals","input":{"period":"rolling_30d"}}

dispatch.indexPublic

Return the catalog of 8 dispatch trade verticals. Per trade: the BLS OES hourly WAGE (medianRateDollars/p25/p75 — what a technician earns, NOT a price) alongside billedHourlyRateRoutineDollars, what a buyer pays hourly at routine urgency, derived by the same wage→billed transform dispatch.estimate uses. See the wage_vs_billed block.

{"skill":"dispatch.index","input":{}}

dispatch.quotex-api-key

Return an instant dispatch quote for a field-service job: accepts trade + location + urgency, returns matched contractor availability, estimated cost, ETA tier, and quote validity.

{"skill":"dispatch.quote","input":{"trade":"plumbing","location":{"state":"MO"},"urgency":"routine"}}

dispatch.orderx-api-key

Create a tracked dispatch work order. Auto-matches a contractor from the active pool and returns a masked contractor reference, order id, and status URL.

{"skill":"dispatch.order","input":{"trade":"electrical","location":{"state":"AZ"},"description":"breaker panel fault"}}

▸ Live response preview

dispatch.evals returns canonical metrics from production telemetry. Numbers carry an estimated flag — see evals.estimated: true in the full response.

GET /api/public/dispatch-metricsfetching...

Get API access.

API keys are issued manually — talk to us and we’ll get you a live key and sandbox access within the day.

Get API Access Dispatch Engine overview

Live evals JSON ↗Evals dashboard ↗Trust & transparency ↗Agent discovery ↗