Use SoWhatify from any AI
SoWhatify is a remote MCP server. Any MCP client can connect to it and use your lenses, chains and personas on whatever you are reading.
- Build private lenses, chains and personas in conversation.
- Browse the public marketplace for inspiration.
- Run
run_lensorrun_chainon any URL or pasted text. - Publish on the website when you want to share.
- Triage what people send to your Dropboxes.
Dropboxes are pages you send to someone without an
account: they upload a document, it runs through one of your lenses, and you both
get the analysis. Your client can read them with list_dropboxes and
list_dropbox_submissions and talk you through what came in. Creating
one, changing its provider key or switching it off stays on the website — each is
a public-facing act involving a credential.
Your client is the model — SoWhatify supplies the perspective, the profile slice and the structure. Nothing you make here is public until you publish it yourself.
Connect
Most clients need no token. Claude, Cursor and Zed sign you in through
your browser: you approve once, and the client keeps its own key after that. Leave every
token and header box empty.
You only need a token if your client has no sign-in step — a config file, or an older
IDE. Then you paste your personal token into an Authorization header.
So: if a client offers to sign you in, let it. Copying a token in by hand is the long way
round, and easy to get wrong.
Claude Desktop connects with just the URL below — it opens an authorization page in your browser, and approving there is what gives it a credential. Leave every token and header field empty.
- Settings → Connectors → Add custom connector
- Paste the URL below and name it (anything you like)
- Authentication: leave "Sign in now" selected
- OAuth client: leave "Use Claude's published identity" selected
- Leave Request headers empty, click Add, then approve in the browser
- Nothing goes in Request headers. That box is for servers that want an API key; here the approval page hands Claude its own credential and Claude stores it.
- Claude may show "No sign-in — Detected" next to the third option. Ignore it: it comes from one open endpoint used for discovery, and sign-in works.
- No account needed. The approval page can create one for you with no email, and you can attach an email later without losing anything.
- Use the token below only if you pick "No sign-in" on purpose, or use the older config file — everything else is tokenless.
Prefer a token, or an older build without the Connectors panel?
Two token routes, both equivalent. In the Connectors dialog: pick "No sign-in", then Add header with name Authorization and value "Bearer <your token>". Or, on an older build: Settings → Developer → Edit Config, add this to claude_desktop_config.json, restart. Either way you are replacing the browser sign-in, not adding to it.
Claude Code connects without a token — it opens an authorization page in your browser, and approving there is what gives it a credential. Leave every token and header field empty.
- Run the command below in any project
- Run /mcp and authorize in the browser
- No token in the command: /mcp opens an authorization page the first time you connect.
- There is also a plugin that bundles a skill explaining how to use the tools well.
Prefer a token, or on a machine with no browser?
Sends your personal token as a header instead of running the browser sign-in.
Claude.ai (web) connects with just the URL below — it opens an authorization page in your browser, and approving there is what gives it a credential. Leave every token and header field empty.
- Settings → Connectors → Add custom connector
- Paste the URL below and name it (anything you like)
- Authentication: leave "Sign in now" selected
- OAuth client: leave "Use Claude's published identity" selected
- Leave Request headers empty, click Add, then approve the permissions
- Nothing goes in Request headers, and there is no token to paste anywhere: approving is what gives Claude its credential.
- If Claude labels the "No sign-in" option as Detected, ignore it — sign-in works.
- No account needed. The approval page can create one for you with no email.
- To use a personal token instead, pick "No sign-in" and add a header named Authorization with the value "Bearer <your token>". That replaces the sign-in rather than adding to it.
ChatGPT connects with just the URL below — it opens an authorization page in your browser, and approving there is what gives it a credential. Leave every token and header field empty.
- Settings → Connectors → Create (developer mode may need enabling first)
- Paste the URL below and choose OAuth authentication
- Approve the permissions
- Deep Research mode will not see this server: it requires a specific search/fetch tool pair that this server does not implement.
Cursor has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Settings → MCP → Add custom MCP server
- Choose Streamable HTTP
- Paste the block below into mcp.json
VS Code / Copilot has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Create .vscode/mcp.json in your workspace
- Add the block below
- VS Code uses "servers" rather than "mcpServers".
Windsurf has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Cascade → MCP servers → Configure
- Add the block below
- Windsurf calls the field "serverUrl".
Zed connects with just the URL below — it opens an authorization page in your browser, and approving there is what gives it a credential. Leave every token and header field empty.
- Settings → AI → MCP Servers → Add Remote Server
- Paste the URL below
- Zed offers to authorize in the browser when no token is supplied.
Gemini CLI has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Add the block below to ~/.gemini/settings.json
- Use "httpUrl", not "url" — "url" selects the deprecated SSE transport, which this server does not serve.
LM Studio has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Program → Install → Edit mcp.json
- Add the block below
LibreChat has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Add the block below to librechat.yaml
- Each user supplies their own token in the UI
- customUserVars is the cleanest per-user token story of any client — one server entry serves everybody.
n8n has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Add an MCP Client Tool node
- Set the transport to HTTP Streamable and paste the URL
- Add a Bearer Auth credential holding your token
Any other MCP client has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Add a remote MCP server using the Streamable HTTP transport
- Paste the URL below
- Send your token as the header: Authorization: Bearer <token>
- Protocol revisions 2024-11-05 through 2026-07-28 are all supported, so a client pinned to any of them works.
- There is no SSE stream — the endpoint answers POST only.
A client that only speaks stdio has no sign-in step, so
this one does need a token: the server URL plus your personal token, sent as an
Authorization header. Your token lives on the account page, under Connections.
- Add the block below to your client's MCP config
- The header argument has no space after the colon on purpose: mcp-remote splits its arguments on spaces and would otherwise mangle the value.
- --transport http-only stops it probing for an SSE endpoint that does not exist, so a failure is immediate rather than a hang.
- npx downloads on every launch, so raise your client's startup timeout if the first run times out.
Do you need a token?
Only if your client has no sign-in step. Claude (desktop, web and Code), Cursor and
Zed all open an authorization page instead — approve it once and they hold
their own credential, so every token and header field in their setup dialog stays
empty. A config file, or a client you deliberately set to “no sign-in”,
needs your personal token as an Authorization header:
No account? Connect anyway
No sign-up either way. A client that signs in through the browser offers this on the approval page itself — “Continue without an account” — so you need nothing from here. The button below is the token version of the same thing, for a client that cannot sign in. Either way you get a private id and can build lenses, chains and personas (never publish) with generous limits. Add an email later to publish and sync across devices — your id and everything you built are kept.
Send it as Authorization: Bearer <token>. Only a hash is stored, so
this is the one time it can be shown. Connecting again replaces this token.