Connect AI clients to your workspace

Zanfia exposes a hosted Model Context Protocol server. Add it to any MCP-capable AI client — Claude Code, Claude Desktop, Cursor, VS Code, Codex — and the assistant gets typed tools to read and act on your workspace: list products, write course lessons, create offers, post in your community, pull sales stats.

Connect my Zanfia workspace as an MCP server. Zanfia = platform for
selling digital products (courses, communities, pages).

Before configuring, ask me, one at a time, and wait for answers:

1. Which client? (options: "claude-code", "cursor", "vscode",
   "claude-desktop", "codex")
2. My Zanfia API key? (I create it at Dashboard → Integrations →
   API, MCP and CLI;
   read-only scopes for analysis, write scopes only if the assistant
   should create content)

Server: type "http" (Streamable HTTP), URL https://app.zanfia.com/mcp,
header "Authorization: Bearer <API_KEY>".

claude-code → run:
  claude mcp add zanfia --transport http https://app.zanfia.com/mcp \
    --header "Authorization: Bearer <API_KEY>"
codex → add to ~/.codex/config.toml (and export ZANFIA_API_KEY):
  [mcp_servers.zanfia]
  url = "https://app.zanfia.com/mcp"
  bearer_token_env_var = "ZANFIA_API_KEY"
other clients → add to the MCP config JSON:
  { "mcpServers": { "zanfia": { "type": "http",
    "url": "https://app.zanfia.com/mcp",
    "headers": { "Authorization": "Bearer <API_KEY>" } } } }

Verify by calling the list-products tool. Tools follow the key's scopes:
*:read for reads, *:write for writes; insufficient-scope = the key lacks
that permission. Tool list: https://zanfia.com/help/developer-mcp/connect.md
For anything the tools don't cover, the full REST API spec is at
https://zanfia.com/openapi/workspace-api.json (docs: https://zanfia.com/llms.txt).

Authentication

The server authenticates with the same bearer tokens as the API: create a scoped API key and send it as an Authorization: Bearer header. Tool access follows the key's scopes — read tools need the matching *:read scope, write tools *:write.

Add it to your client

Claude Code:

claude mcp add zanfia --transport http https://app.zanfia.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Cursor / VS Code / Claude Desktop (JSON MCP config):

{
  "mcpServers": {
    "zanfia": {
      "type": "http",
      "url": "https://app.zanfia.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Codex (OpenAI Codex CLI) — add to ~/.codex/config.toml (or a project-scoped .codex/config.toml), with the API key exported as an environment variable:

[mcp_servers.zanfia]
url = "https://app.zanfia.com/mcp"
bearer_token_env_var = "ZANFIA_API_KEY"

export ZANFIA_API_KEY="YOUR_API_KEY"

You can also manage the entry from the terminal with the codex mcp subcommand (codex mcp list, codex mcp add, …) instead of editing the TOML by hand.

Available tools

The server exposes the full Workspace API surface as ~90 tools. Read tools require the matching *:read scope, write tools *:write — products, offers, courses, media, clients, orders, stats, and discounts sit under the products scope; pages and communities under their own.

Destructive tools (deletes, access revocation) say so in their descriptions, so a well-behaved assistant will confirm before using them. File and video uploads are deliberately not MCP tools — binary payloads don't fit tool calls; use the CLI or API for uploads, then reference the media id from MCP.

Community members

The same endpoint also accepts community-member API keys (created by a member of your community in the member platform under Settings → Developer). A member key gets a separate, member-scoped tool set — the MCP twin of the Community API: listing their communities and channels, reading and searching posts, comments and chat, writing posts/comments/messages as themselves (own content only), and raising and tracking their support tickets. None of the workspace tools above are available to member keys.

For agents that keep an eye on a community, the get-community-updates tool (also GET /v1/communities/:communityId/updates?since=<ISO> on the Community API) returns everything new — posts, comments, and chat — across the caller's readable channels in one call: pass the returned nextSince as the next since instead of polling each channel separately. To know when someone talks to you (replies, @-mentions, reactions, DMs), member keys also get get-my-notifications and mark-notifications-read — the member's own notification inbox, shared with the platform bell.

Member requests are rate limited (per minute, per member — a 429 carries Retry-After), and content creates accept a clientRequestId UUID so retries never duplicate a post, comment, or message.

Member access is off by default: it works only in communities where the creator has enabled member API access in the community settings, and flipping the setting off immediately locks out already-issued member keys.

MCP, CLI, or raw API?

• MCP — best when a person chats with an AI client and wants it to act on the workspace directly.
• CLI — best for terminal-based agents, scripts, and CI; also the route for file and video uploads, which MCP deliberately doesn't do.
• Raw API — best for your own integrations and backends.

All three are the same underlying platform with the same permissions model, so you can mix them freely.