Why do I get a 401 error?

Sign in to Sesavu again through your client. If you use a personal access token, check the Authorization: Bearer header and replace an expired or revoked token in Sesavu → Agents.

Why do I get a 403 or insufficient_scope error when saving?

Read only access cannot save. Configure your client to request both mcp:read and mcp:write. Disconnect the connection in Sesavu → Agents, then reconnect and choose Read and save. If that choice is missing, re-add or reconfigure the server in your client with both scopes. Your client or workspace must also allow write tools.

Why does Sesavu say waiting for first read?

Signing in or listing tools does not prove that your assistant has read your library. Enable Sesavu in your conversation and ask: List my collections in Sesavu. A successful read from that client establishes the connection; the browser diagnostic does not count.

What should I do after a 429 error?

Wait for the number of seconds in the Retry-After response header, then try again.

Documentation / Agents

Connect your AI
to Sesavu.

Give your AI assistant access to your Sesavu reference library through the Model Context Protocol (MCP). Search references, read collections and stored design briefs, or choose permission to save links and organize collections. Agent access is free.

Copy the MCP server URL

https://sesavu.com/api/mcp

Transport: Streamable HTTP. Use this same endpoint in every client.

Sign in or use a personal access token

Sign in with Sesavu (OAuth). Add the server in your client, then sign in to your Sesavu account in the browser and approve access.

Choose Read only (mcp:read) or, if your client requests both scopes, Read and save (mcp:read and mcp:write). Start with Read only if your assistant only needs context.

Personal access token (PAT). For clients that accept bearer headers, open Sesavu → Agents, select your client, and choose the personal access token option. Choose the scope and create a token. Copy it into your client’s private credential settings:

Authorization: Bearer <your token>

Keep tokens out of URLs, shared files, and repositories. Claude and ChatGPT setup below uses browser sign-in; do not paste a token into chat.

OAuth clients discover sign-in at https://sesavu.com/.well-known/oauth-protected-resource/api/mcp.

Connect your client

Connect Claude

Open Customize → Connectors → Add custom connector.
Name it Sesavu and paste the server URL above.
Add the connector, then connect and sign in to Sesavu. Approve the access you want.
Enable Sesavu in your conversation and ask: List my collections in Sesavu.

Workspace permissions can control who adds connectors. Claude’s connector instructions.

Connect ChatGPT

On ChatGPT web, open Plugins → + → Add custom MCP server.
Name it Sesavu. Enter the server URL under Connection and choose OAuth authentication.
Review the warning, acknowledge it, and choose Create as a plugin. Sign in to Sesavu and approve access when prompted.
Install the plugin, start a conversation, and select Sesavu with @. Ask: List my collections in Sesavu.

Saving through a custom MCP server requires a Business, Enterprise, or Edu plan and workspace permission. Pro supports read/fetch connections. Custom server setup is web-only.

Business setup requires an admin or owner; Enterprise/Edu can authorize members. ChatGPT’s custom MCP requirements.

Connect Claude Code

Run this command in your terminal:

claude mcp add --transport http --scope user sesavu https://sesavu.com/api/mcp

In Claude Code, run /mcp, choose sesavu → Authenticate, and sign in through your browser. Approve access, then ask Claude Code to list your Sesavu collections.

Connect Codex

Run these commands in your terminal:

codex mcp add sesavu --url https://sesavu.com/api/mcp
codex mcp login sesavu

Approve access in the browser. Ask Codex to list your Sesavu collections. The Codex IDE extension shares this MCP configuration.

Connect Cursor

Open Add to Cursor on the computer where Cursor is installed.
Review and install the server configuration.
Sign in to Sesavu when prompted and approve access. Ask Cursor to list your Sesavu collections.

Connect VS Code / GitHub Copilot

Open Add to VS Code on the computer where VS Code is installed.
Review the configuration and trust/start the server.
Sign in to Sesavu and approve access. Invoke Sesavu in Copilot Chat and ask it to list your collections.

Alternatively, run MCP: Add Server → HTTP from the Command Palette and paste the server URL above.

What tools can my agent use?

Seven tools. Write tools require Read and save access.
Tool name	What it does	Access
`conext_search_references`	Search your saved references, with up to 50 results per call.	Read
`conext_get_reference`	Read one reference, its metadata, and short-lived media links. Captured text is marked as untrusted.	Read
`conext_get_design_brief`	Read a collection’s stored design brief and design tokens, when available. This does not generate a brief.	Read
`conext_list_collections`	List your collections, with up to 100 results per call.	Read
`conext_save_reference`	Save a webpage link with an optional title, notes, and collection. This does not fetch or capture the page.	Write
`conext_add_to_collection`	Add one of your references to one of your collections. Repeating the same addition does not duplicate it.	Write
`conext_create_collection`	Create a private collection with a title and a unique lowercase, hyphen-separated slug.	Write

The conext_ tool IDs keep existing connections compatible. The server introduces itself as Sesavu.

Understand limits and privacy

Agent access is free. Normal save and storage limits apply.

The connection can read your account’s references and collections, including private material. Read and save also allows the three write tools above. There is no delete tool, and the connection cannot access another person’s private library.

Content returned by a tool is shared with the AI client you connect. Review that client’s data settings. Reference text is marked as untrusted content; it is source material, not instructions for your assistant.

Read requests: 60 per minute and 1,000 per hour.
Write requests: 30 per minute and 300 per hour.
Design brief reads: 20 per hour.
Responses are capped at 256 KiB and may be truncated.

Rate limits apply per credential. OAuth connections also share limits across your account.

Disconnect an agent

Open Sesavu → Agents, choose Connected, then Disconnect for the connection. This revokes its access. Remove the server in your client too if you no longer want it listed. You can revoke personal access tokens from the same page.

Fix connection problems

Why do I get a 401 error?: Sign in to Sesavu again through your client. If you use a personal access token, check the Authorization: Bearer header and replace an expired or revoked token in Sesavu → Agents.
Why do I get a 403 or insufficient_scope error when saving?: Read only access cannot save. Configure your client to request both mcp:read and mcp:write. Disconnect the connection in Sesavu → Agents, then reconnect and choose Read and save. If that choice is missing, re-add or reconfigure the server in your client with both scopes. Your client or workspace must also allow write tools.
Why does Sesavu say waiting for first read?: Signing in or listing tools does not prove that your assistant has read your library. Enable Sesavu in your conversation and ask: List my collections in Sesavu. A successful read from that client establishes the connection; the browser diagnostic does not count.
What should I do after a 429 error?: Wait for the number of seconds in the Retry-After response header, then try again.

Still stuck? Contact support.