Mythweaver runs a Model Context Protocol (MCP) server, so outside AI clients such as Claude, Cursor or Codex can use the same tools as Chat with Mythweaver: browsing and editing your characters, lore, prompts, chats and settings. Clients authenticate with a personal access token that acts as you.
Before you start
You need the Connect AI tools over MCP permission (mcp.use). If an administrator has turned it off for your account, the MCP access section says MCP access is turned off for this account by an administrator. and lists your existing tokens only so you can revoke them. See Users, groups and permissions.
To create a token
- Open your account settings at
/users/settingsand find MCP access. - Enter a Token name (up to 80 characters, unique among your tokens), for example the client's name.
- Pick the Access level.
- Click Create token. Because a token is a long-lived credential, you may be asked to confirm your password first (sudo mode; see Your account).
- Copy the secret that appears. It starts with
mwp_and is shown only once: Mythweaver stores only a fingerprint of it. Click Done when you have saved it.
Access levels
| Access | What the token can do |
|---|---|
| Read only | Only the reading tools: list, get and search. |
| Full access | Every MCP tool, including writes, with your permissions except site administration and moderation. |
| Full access, including site administration | Everything Full access has plus your administration and moderation permissions. Offered only if you are the owner or hold a site permission. |
Warning
Write tools run as soon as the client calls them. Unlike Chat with Mythweaver, there is no Approve/Reject step, so give write access only to clients you trust.
To connect a client
Configure the client with:
- the endpoint: your server's address followed by
/mcp, for examplehttps://mythweaver.example.com/mcp; - the header
Authorization: Bearer <token>, with the secret you copied.
The endpoint speaks MCP's Streamable HTTP transport in stateless mode: clients send JSON-RPC requests with POST and get JSON replies. GET and DELETE return 405, since there is no server-to-client stream or session. A client that asks for a JSON configuration typically looks like this:
{
"mcpServers": {
"mythweaver": {
"type": "http",
"url": "https://mythweaver.example.com/mcp",
"headers": { "Authorization": "Bearer mwp_…" }
}
}
}
The exact format depends on the client; check its documentation.
To see what a token has done
Each token in the list shows its access level, the last four characters of its secret, when it was created and when it was last used. Expand Recent calls to see its tool calls from the last 30 days (up to 20 per token), each marked as succeeded or failed.
To revoke a token
Click Revoke next to the token and confirm. Clients using it stop working immediately. Revoking doesn't need sudo.
How tokens are checked
- Every request re-checks your current permissions, so removing
mcp.usefrom you stops all your tokens on their next call. - A token never grants sudo, so actions that need you to confirm your password in the app can't be done over MCP.
- A suspended account, or a request from a banned device or network, is refused. Each request records its IP address and user agent among the account's sign-in fingerprints; see Moderation.
| Response | Meaning |
|---|---|
| 401 Missing or invalid MCP access token. | No token, a mistyped one, or a revoked one. |
| 403 MCP access is turned off for this account. | You no longer hold mcp.use. |
| 403 This account or network is blocked. | The account, device or network is banned. |
Available tools
MCP clients get the assistant's tool catalog, minus a few tools that only make sense inside a Mythweaver conversation or that need sudo: the API key form, reading a stored tool result, and moving lore books, connections or models to the site. A read-only token sees only the reading tools. The tools cover characters and their outfits, sprites and expressions; lore books, entries and pools; chats, participants, messages and injections; AI connections and models; prompts, blocks and outlets; settings, variables, scripts, memory settings, jobs, analytics, tags and search; and this documentation (list, search and read its pages). Each tool describes its own arguments to the client, and they act with exactly your permissions.