Intercom is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Intercom API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 64/100Comprehension
55% of score · 91/100What we found
- An agent that lands on an Intercom endpoint page gets what it needs immediately: each reference page leads with a task-oriented sentence ("You can create a new contact (ie. user or lead)") followed by the method, path, version, and security scheme before any detail — answer-first by default.
- Every documentation page is available as clean, structured markdown via a `.md` suffix, plus a full llms.txt index — so an agent never has to scrape JS-rendered HTML to read the docs.
- A complete, versioned OpenAPI 3.0.1 spec is public on GitHub (intercom/Intercom-OpenAPI) with raw YAML per version (2.7–2.15), giving an agent a machine-readable contract for the entire surface.
- Discovery is held back by basics: there is NO robots.txt at all (returns 404), no llms-full.txt, and no /.well-known/mcp.json — so the site sends no explicit crawler signals even though the content underneath is agent-friendly.
- The llms.txt exists and is developer-focused, but at ~243KB it is far too large to load as a single agent context window — it is a flat link dump of every doc page across every API version rather than a right-sized summary.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add a robots.txt that explicitly allows AI crawlers (GPTBot, ClaudeBot, Google-Extended, CCBot) — the absence of any robots.txt is the single biggest discovery gap and a trivial fix.
- 02Publish a right-sized llms.txt (<50KB) scoped to the current stable API version with descriptive sections, and move the exhaustive all-versions index to llms-full.txt — the current 243KB file is too large for agents to consume in one pass.
- 03Add a standardized AGENTS.md to the primary public repos (the OpenAPI repo already has a useful CLAUDE.md — promote/duplicate it as AGENTS.md so it matches the rising cross-tool convention).
- 04Expand error documentation from a code-and-description table to explicit recovery guidance for the top 5–10 error scenarios (e.g. what to send on 409 conflict, how to back off on 429 beyond the existing rate-limit section).
- 05Publish a /.well-known/mcp.json pointing at Intercom's already-live remote MCP server so agents can auto-discover it from the docs domain.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Execution characteristics are documented thoroughly: Bearer access-token and OAuth2 auth, machine-parseable JSON error objects (error.list with stable errors[].code values and a request_id), explicit rate limits (10k/min per app, 25k/min per workspace) with X-RateLimit-* headers, and cursor-based pagination. No idempotency-key support is documented.