Basic Machines is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Basic Machines 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 · 71/100Comprehension
55% of score · 96/100What we found
- An agent looking for Basic Machines' documentation lands first on the basicmachines.co marketing splash page, which has no robots.txt, llms.txt, sitemap, or .well-known/mcp.json — every discovery signal 404s. The real, agent-ready docs live on a separate domain (docs.basicmemory.com), reachable only by following a body link.
- Once on docs.basicmemory.com, comprehension is exceptional: a well-structured llms.txt index, a full llms-full.txt dump, clean markdown via content negotiation (/raw/*.md), plus a dedicated 'AI Assistant Guide' and 'MCP Tools Reference' written specifically for AI agents.
- This is a genuinely agent-native product — it ships an AGENTS.md in its main repo, pre-built Agent Skills, and is listed on PulseMCP (official), Glama, and Smithery — so an agent that finds the GitHub repo or an MCP registry discovers it easily.
- Because Basic Memory is MCP-native rather than a REST API, it has no OpenAPI/Swagger spec and no .well-known/mcp.json — the two discovery signals it fails — which caps the Discovery score even though its docs are outstanding.
- Every MCP tool is documented task-first ('Create or update a markdown note', 'Build context graph from a memory URL') with realistic, copy-pasteable examples, consistent snake_case naming, and explicit recovery guidance for the common errors an agent will hit (note-already-exists, missing project, entity-not-found).
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add discovery files to the primary domain: publish robots.txt, sitemap.xml, and an llms.txt on basicmachines.co that points agents to docs.basicmemory.com — right now the canonical company domain gives an agent nothing to work with.
- 02Publish a .well-known/mcp.json on docs.basicmemory.com (and/or basicmachines.co) declaring the Basic Memory MCP server, its tools, and auth — this is the emerging standard for MCP discovery and is the single cheapest Discovery win available.
- 03Add a dedicated capability-boundaries / limits page: consolidate Cloud rate limits, traversal-depth performance caveats, and feature-availability differences between local and cloud into one place so agents learn boundaries before failing.
- 04Cross-link the domains explicitly for machines: reference docs.basicmemory.com from a basicmachines.co llms.txt and vice-versa, so an agent starting from either the marketing site or the docs site can traverse the full surface.
- 05Consider exposing an OpenAPI spec for the Basic Memory Cloud REST/API-key surface (api-keys, workspaces, snapshots) so integrators building outside MCP have a machine-readable contract.
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.
Basic Machines' flagship product Basic Memory is an MCP server, so the agent-facing 'API' is a set of MCP tools rather than REST endpoints. Tool parameters, machine-readable JSON output (output_format="json"), pagination (page/page_size), and a dedicated error-handling/recovery section are all documented. Auth is OAuth or API keys for Basic Memory Cloud and filesystem-local (no auth) for local mode. Rate limits and idempotency keys are not explicitly documented.