ThinkClear MCP server
One endpoint, Streamable HTTP, OAuth 2.1. There is no API key to request and nothing to paste — the server registers your client itself and issues the token on first call.
The endpoint
https://app.thinkclear.xyz/api/mcpStreamable HTTP transport, served statelessly: every request builds a fresh server from the token it carries, which is what lets the token decide both the owner and the tool list. Clients speaking the 2025-era protocol revisions are accepted as well as current ones.
Claude Code
claude mcp add --transport http thinkclear https://app.thinkclear.xyz/api/mcpAuthorization
An unauthenticated call is answered with 401 and an RFC 9728 challenge naming the protected resource metadata, which names the authorization server, which advertises dynamic client registration. Your client walks that chain on its own: it registers itself, opens a browser, you approve the scopes on a consent screen, and it holds a token it refreshes without asking again. Grants are listed in the app's account screen and revoked from there.
- Protected resource metadata — RFC 9728 — what the endpoint is and which server authorizes it
- Authorization server metadata — RFC 8414 — endpoints, scopes, and the registration URL
- Setup guide, per client — inside the app, once you are signed in
Both documents carry a path because the resource and the issuer do. There is nothing at the bare /.well-known/oauth-authorization-server, and that is correct rather than missing: this server's issuer is app.thinkclear.xyz/api/auth, and metadata served at the origin root would claim an issuer that does not match where it was fetched from.
Tools
Twelve, and they are the same objects the built-in assistant calls — one definition, two transports, so anything the chat panel can do to a mindmap a connected agent can do too. Prefer the batch calls: add_topics takes a nested tree and the rename, move and delete tools take lists, so a reorganization is one round trip rather than twenty.
list_mindmapsmindmaps:read- Every mindmap the signed-in user owns.
read_mindmapmindmaps:read- One mindmap as an indented outline, topic ids included.
search_topicsmindmaps:read- Find topics by words in their title or note, across every map.
read_topic_notemindmaps:read- The markdown note on one topic.
create_mindmapmindmaps:write- Start a new map from a title.
rename_mindmapmindmaps:write- Retitle a map and its root topic.
delete_mindmapmindmaps:write- Delete a map. Irreversible.
add_topicsmindmaps:write- Graft a nested tree of topics under a parent, in one call.
rename_topicsmindmaps:write- Retitle topics in a batch.
move_topicsmindmaps:write- Re-parent topics in a batch.
delete_topicsmindmaps:write- Delete topics and their descendants.
set_topic_notemindmaps:write- Replace a topic's markdown note wholesale.
A tool that returns an error also returns the issues that caused it, which is a repair instruction rather than a refusal — fix the edit and call again instead of asking the user what to do.
Scopes
- mindmaps:read — read your mindmaps and the topics inside them.
- mindmaps:write — create, rename, reorganize, and delete your mindmaps and topics.
Ask for the narrower one when the job only reads. Scopes are enforced by leaving tools out of the registration, so a read-only token is served a tool list with no way to edit anything and the agent never plans a call that was going to be refused.