RXed AI News

AI to the bone.
SCORE 7/10 10/10 criteria · Strong Assessed 2026-09-22 · ARI rubric v1.0

Intercom — can an agent use it?

Visit intercom.com
“A complete system for human and AI customer service, combining the Intercom Helpdesk with Fin AI Agent. Used by 30,000+ businesses.” — the vendor’s own words

Intercom scores well on discoverability and documentation, with an official MCP server and explicit AI-welcome signals. The MCP is read-heavy — agents can query conversations and contacts but cannot create tickets or send messages via protocol. No agent payment path exists.

Category
Customer support
Agent protocol
Official
Agent payment rails
None published
Site Scan
n/a (None/100) · crawl-level, 22/09
Tap or hover any criterion to see why it scored that.
FIND & RETRIEVE8.3/10Can an agent get correct information?

Machine-readable surface9/10

Comprehensive llms.txt with structured site map, canonical policy and retrieval rules. Sitemap index with 7 sub-sitemaps. robots.txt includes Content-Signal: ai-train=yes, ai-input=yes, search=yes — one of the most explicitly AI-welcoming signals in the index. source

Information retrieval quality7.5/10

Pricing clearly published: $19/$85/$132 per seat/mo plus $0.99 per Fin AI resolution. Plan comparison table with feature breakdown. 800+ help center articles. API-specific pricing is not surfaced (the platform charges per seat, not per API call). source

Documentation for machines8.5/10

Developer platform at developers.intercom.com with versioned REST API (stable 2.12), OpenAPI spec published on GitHub, regional endpoints documented (US/EU/AU), SDKs, guides and tutorials. Free development workspace available. source

ACT & TRANSACT6.4/10Can an agent do the job?

Action coverage8/10

REST API covers conversations, contacts, companies, tickets, articles, messages, data events, tags, segments, admins, data export, workflows, and more — both read and write. Core helpdesk actions (create contact, manage conversations, resolve tickets) are all API-accessible. source

Agent protocol support7/10

Official MCP server at mcp.intercom.com/mcp with 14 tools, OAuth + bearer auth. Covers conversations, contacts, companies, and help center articles. However, the MCP is primarily read-oriented: agents can search and retrieve conversations but cannot create tickets, send replies, or close conversations via protocol. Article creation and internal notes are the only write paths. source

Access & auth friction8/10

OAuth 2.0 with PKCE for public apps, bearer tokens for private apps. Free development workspace for testing. Self-serve credential generation via Developer Hub. Scoped permissions per app. source

Agent payment capability1.5/10

No evidence of ACP, AP2, MPP, x402 or any agent payment protocol. Subscriptions are managed through standard billing pages. An agent cannot programmatically subscribe, upgrade, or pay on behalf of a customer. source

Cost & rate fairness7.5/10

Rate limits are 10,000 API calls/min per app and 25,000/min per workspace — generous for an SMB helpdesk integration. Higher limits available on request. No per-call pricing for the API itself (included in seat cost). No evidence of agent-vs-human price discrimination. source

TRUST & DURABILITY8.2/10Will it still work next quarter?

Permission & ToS stance8.5/10

robots.txt explicitly includes Content-Signal: ai-train=yes, ai-input=yes, search=yes — among the strongest AI-welcome signals seen. Terms section 2.5 permits API access with documented rate limits. No anti-automation clause found. llms.txt actively guides LLMs to canonical sources. source

Reliability & continuity8/10

Regional status pages at finstatus.com for US/EU/AU with real-time dashboards. API versioned (2.12 stable) with documented deprecation policy. OpenAPI spec maintained on GitHub. Long track record of backward compatibility across major versions. 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.