Felt Maps

Yes — the Enterprise plan. Felt's MCP documentation states the feature is only available to customers on the Enterprise plan, and Felt's billing page lists AI features and MCP among what Enterprise adds over Professional. Felt's AI assistant FAQ separately describes Felt AI as an add-on on Professional, so confirm MCP access with Felt before assuming a Professional workspace qualifies.

No. Felt's own OpenAPI specification for the equivalent REST endpoint states the action cannot be undone, and that the map and all its layers, elements, and comments will be permanently removed. No restore, undelete or trash tool appears in either published tool list. Felt's disaster-recovery language covers infrastructure continuity, not reversing a user-requested deletion.

Yes. Felt documents share_map as setting a map's public access level and returning its share link. Felt's REST API publishes the matching enum of four values: private, view_only, view_and_comment, and view_comment_and_edit. Three of those four grant access to anyone holding the link, so an agent calling this tool can publish geographic data.

Yes. Felt's OAuth metadata advertises five scopes: map.read, map.write, layer.read, layer.write and profile.read. We fetched both the resource descriptor and the authorization server metadata on 2026-08-18 and both list the same five. That per-resource split means a consent screen can offer read-only, or map access while withholding layer writes.

Felt documents a render_map tool that renders a map inline as an interactive widget, described as a one-shot snapshot of the map's state when called. That tool appears in Felt's own reference but not in Anthropic's directory snapshot of 2026-08-16, so we cannot confirm which clients receive it. We never authenticated or called it.

Felt's developer documentation publishes two independent limits for its API: a per-IP throttle of 300 requests per minute, and a plan-level usage limit surfaced through an x-api-limit-exceeded response header. Both return HTTP 429 with a too_many_requests code. Felt documents these for its REST API rather than specifically for the MCP endpoint.

Edit permission, plus a Full Access seat. Felt's permissions reference lists Delete map, Delete layers, Upload data, Create new layers from scratch and Spatial analysis as Edit-level actions, unavailable to View or Contribute. Felt states MCP uses your existing workspace permissions, so the connector cannot exceed what your own account may already do.

No. Felt's AI assistant documentation states plainly that Felt does not train models on your data and that none of it becomes public. Felt adds that database connections are read-only unless you grant write access, that generated SQL only ever runs SELECT statements, and that Felt AI runs on frontier models hosted on Amazon Bedrock.