RXed AI News

AI to the bone.
SCORE 6/10 10/10 criteria · Capable, with friction Assessed 2026-08-22 · ARI rubric v1.0

Exact Online — can an agent use it?

Visit exact.com
“Cloud-based accounting and ERP platform for SMEs and their accountants, dominant in the Benelux mid-market with 675,000 users.” — the vendor’s own words

Exact Online has a broad REST API with 200+ endpoints and an active partner ecosystem building MCP integrations, but Exact itself publishes no agent protocol and the tight rate limits and app-review gate add friction for autonomous agents.

Category
Accounting
Agent protocol
Community only
Agent payment rails
None published
Site Scan
C (66/100) · crawl-level, 22/08
Sf5.5
Machine-readable surface
Iq4.5
Information retrieval quality
Dx6.5
Documentation for machines
Ac7.5
Action coverage
Pt4.5
Agent protocol support
Au6
Access & auth friction
Py2
Agent payment capability
Cf5.5
Cost & rate fairness
Ts7
Permission & ToS stance
Rl6.5
Reliability & continuity
Tap or hover any criterion to see why it scored that.
FIND & RETRIEVE5.5/10Can an agent get correct information?

Machine-readable surface5.5/10

Valid multi-country sitemap referenced in robots.txt, clean server-rendered HTML (no JS-wall), and robots.txt does not block AI crawlers. However, no llms.txt, no RSS/JSON feeds, no markdown negotiation, and only BreadcrumbList in schema.org — an agent can crawl but finds limited structured signals. source

Information retrieval quality4.5/10

Pricing is scattered across country-specific pages (US $25–$739/mo by module) with no unified machine-readable price list; the main /pricing path returns 404. Terms and conditions are PDF downloads behind a country selector. API documentation and rate-limit details sit behind the support portal login at support.exactonline.com. An agent can find product marketing copy easily but would struggle to extract specific prices, limits, or contractual terms. source

Documentation for machines6.5/10

Comprehensive REST API reference at start.exactonline.nl/docs with 200+ endpoints across 30+ service categories, OAuth 2.0 flow documented, code examples provided, division-based architecture explained, and rate limits published. Docked because there is no public API changelog or versioning history, no OpenAPI/Swagger spec found, and some developer docs require a support portal login. source

ACT & TRANSACT5.1/10Can an agent do the job?

Action coverage7.5/10

Over 200 REST API endpoints covering accounting, CRM, sales, purchase, inventory, logistics, HRM, projects, and manufacturing with full CRUD on most transactional entities. Bulk endpoints support 1,000 records per call, and webhooks are available for event notifications. Notable gap: invoice-payment reconciliation (a core accounting action) has no REST endpoint and requires the legacy XML API with manual file uploads. source

Agent protocol support4.5/10

Exact publishes no official MCP server, WebMCP, or agent protocol. Not in the official MCP registry (modelcontextprotocol/servers). Multiple community and partner-built MCP servers exist: iWebDevelopment (an official Exact partner) offers a hosted remote MCP server with read/write access to GL, invoices, orders, and stock; Peliqan publishes an open-source MCP server; Apideck and Zapier offer normalized MCP layers. The iWebDevelopment server is comprehensive but is partner-maintained, not Exact-published. source

Access & auth friction6/10

OAuth 2.0 with self-serve app registration via the Exact Online App Center. However, apps are sandboxed to the creating tenant until Exact manually reviews and approves them — a significant blocker for multi-tenant or agent scenarios. Token refresh is capped at once per 10 minutes. Every API call requires a division ID, adding an extra discovery step. Scoped credentials are possible but the approval gate adds weeks of friction. source

Agent payment capability2/10

No evidence of any agent payment protocol (ACP, AP2, MPP, x402). Exact Online subscriptions are paid via direct debit through the standard portal. An agent cannot programmatically subscribe, upgrade, or pay on a customer's behalf. The platform processes payments for its users' businesses but offers no delegated spend or agent checkout path for the service itself. source

Cost & rate fairness5.5/10

Base rate limits are 60 calls/minute and 5,000 calls/day per company, documented publicly. Volume packs are available (each adding 5,000 daily calls and 25,000 booking lines). API access is included with every subscription — no separate API tier pricing or agent surcharge. However, the 60/min base limit is tight for heavy automation, pipelining (parallel requests) is explicitly banned, and XML calls count as 50 API calls each. Exceptions can be requested with a justified analysis. source

TRUST & DURABILITY6.8/10Will it still work next quarter?

Permission & ToS stance7/10

API integrations are actively encouraged through the App Center partner ecosystem. robots.txt blocks only Joomla admin paths with no AI-specific disallows. Site conditions prohibit reverse engineering but contain no explicit bot, automation, or scraping restrictions. Fair use policy is reasonable: 'excessive' means significantly exceeding average SME usage, and corrective action starts with notification. No known litigation against integrators. Docked slightly for the vague fair-use language and the app-review gate's chilling effect on unapproved automation. source

Reliability & continuity6.5/10

Public status page at status.exact.com (Atlassian Statuspage) with per-component status, subscribable via email/webhook/Teams/RSS/Atom — strong programmatic transparency. The API has remained on v1 for years with no breaking version changes. However, 1,613 outages tracked over ~4 years (StatusGator), no public API changelog exists, and four APIs were deprecated in June 2023 with communication only through partner channels. Exact is a 40+ year old company (Delft, NL) with 675,000 users, providing institutional stability. source

Sources

Point-in-time assessment — services change terms, prices and APIs often. That volatility is itself scored under Reliability & continuity. Re-verified at least quarterly.