# Connect your AI to ViewsMax ViewsMax exposes an MCP server and a REST API so AI assistants and agents can post to a user's connected social accounts (YouTube, TikTok, X, LinkedIn, Threads, Instagram, Bluesky), manage offers and tracked links, read click/conversion/revenue analytics, and research outlier videos (content that massively over-performed its channel, with AI breakdowns of why) — on the user's behalf, with their permission. User data is private; all access is authenticated. - **MCP endpoint:** `https://api.viewsmax.com/api/mcp` (Streamable HTTP) - **Capability discovery (JSON):** `https://api.viewsmax.com/api/ai` - **REST API:** `https://api.viewsmax.com/api` — [OpenAPI spec](https://api.viewsmax.com/docs.openapi) · [API reference](https://api.viewsmax.com/docs) - **MCP server overview:** (plain text: ) — what it exposes, auth modes, rate limits, and all 30 tools grouped by job Step-by-step guides per agent: [Claude](https://viewsmax.com/claude) · [Claude Code](https://viewsmax.com/claude-code) · [Claude Cowork](https://viewsmax.com/claude-cowork) · [ChatGPT](https://viewsmax.com/chatgpt) · [Codex](https://viewsmax.com/codex) · [Cursor](https://viewsmax.com/cursor) · [OpenClaw](https://viewsmax.com/openclaw) · [Hermes Agent](https://viewsmax.com/hermes). Each has a plain-markdown setup brief at `/.md` (e.g. ). ## Credentials Two options, both sent as a Bearer token: 1. **OAuth (recommended for chat apps).** Point an MCP client at the endpoint above; the user signs in and approves in the browser (OAuth 2.1 + PKCE, dynamic client registration supported). The consent screen offers read-only or full access. No key handling. 2. **API key (for headless agents and scripts).** In the ViewsMax app: **Settings → AI Assistant Access → generate key** (`vmx_...`). Choose read-only or full access. The key is shown once; rotating it invalidates the old one. The same key works on the REST API (posts, offers, tracking, stats, outliers — read-only keys are limited to GET). ## Claude (claude.ai / Claude Desktop / mobile / Cowork) Customize → Connectors → **+** → **Add custom connector** → URL `https://api.viewsmax.com/api/mcp` → Add → Connect and complete the sign-in approval. Free Claude plans can add one custom connector; on Team/Enterprise an owner adds it first under Organization settings → Connectors. ## Claude Code ```bash claude mcp add --transport http --scope user viewsmax https://api.viewsmax.com/api/mcp ``` or in `.mcp.json`: ```json { "mcpServers": { "viewsmax": { "type": "http", "url": "https://api.viewsmax.com/api/mcp" } } } ``` Then run `/mcp` inside Claude Code and sign in. For headless use add `--header "Authorization: Bearer vmx_YOUR_KEY"` (or `"headers": { "Authorization": "Bearer vmx_YOUR_KEY" }` in `.mcp.json`). ## ChatGPT Settings → Security and login → turn on **Developer mode**. Then go to → **+** → paste the MCP URL above and complete OAuth. ViewsMax is an *action* connector (create/schedule posts, read stats) — use it from regular chats with connectors enabled; it is not a deep-research search/fetch source. ## Codex ```bash codex mcp add viewsmax --url https://api.viewsmax.com/api/mcp codex mcp login viewsmax ``` The Codex CLI, IDE extension, and ChatGPT desktop app share this setup. For headless use, set `bearer_token_env_var = "VIEWSMAX_API_KEY"` under `[mcp_servers.viewsmax]` in `~/.codex/config.toml`. ## Cursor `.cursor/mcp.json`: ```json { "mcpServers": { "viewsmax": { "url": "https://api.viewsmax.com/api/mcp", "headers": { "Authorization": "Bearer vmx_YOUR_KEY" } } } } ``` ## OpenClaw Install the ViewsMax skill (SKILL.md drives the REST API): ```bash mkdir -p viewsmax && curl -o viewsmax/SKILL.md https://viewsmax.com/skills/viewsmax/SKILL.md openclaw skills install ./viewsmax --global ``` Then set `VIEWSMAX_API_KEY` (full-access key) under `skills.entries.viewsmax.env` in `~/.openclaw/openclaw.json`. Alternatively point OpenClaw's MCP support at the MCP endpoint above. ## Hermes Agent Add to `~/.hermes/config.yaml` (Hermes opens a browser to sign in on first connection): ```yaml mcp_servers: viewsmax: url: "https://api.viewsmax.com/api/mcp" auth: oauth ``` For a server with no browser, replace `auth: oauth` with `headers: { Authorization: "Bearer vmx_YOUR_KEY" }`. ## Plain REST / curl ```bash curl -H "Authorization: Bearer vmx_YOUR_KEY" https://api.viewsmax.com/api/posts ``` Responses use a `{ success, message, data }` envelope. Full reference: . ## What agents can do (30 MCP tools) `list_connected_accounts`, `list_brands`, `upload_media`, `create_post`, `list_posts`, `get_post`, `update_post`, `delete_post`, `list_offers`, `create_offer`, `get_offer`, `update_offer`, `delete_offer`, `create_tracking_link`, `get_offer_stats`, `get_stats_timeseries`, `disconnect_account`, `get_connect_url`, `create_feature_request`, `list_outliers`, `search_outliers`, `get_outlier`, `fetch_outlier`, `get_outlier_breakdown`, `generate_outlier_breakdown`, `list_saved_outliers`, `save_outlier`, `remove_saved_outlier`, `add_outlier_channel`, `get_outlier_channel_ingest`. Typical posting flow: `list_connected_accounts` → `upload_media` (TikTok / Instagram / YouTube need a video or image) → `create_post` (status `draft`, `posted`, or `scheduled` + `scheduled_at`) → publishing is asynchronous, so poll `get_post` for per-platform results. `create_post`/`update_post` accept an `options` object keyed by platform for per-platform publish settings — e.g. `options.tiktok.privacy_level` and `options.tiktok.auto_add_music` (boolean, default `false`) which lets TikTok auto-add its recommended music to a **photo slideshow** (ignored for video; there is no music option for Instagram). Typical analytics flow: `list_offers` → `get_offer_stats` / `get_stats_timeseries` (clicks + conversions bucketed by hour or day). Typical research flow: `list_outliers` (filter by platform, outlier score, views, subscribers, date, shorts/long, country) or `search_outliers` to scrape a new topic, then `fetch_outlier` for a specific URL → `generate_outlier_breakdown` → poll `get_outlier_breakdown` until `completed` → `save_outlier` with tags to keep it in the user's library. Researching a specific creator: `add_outlier_channel` with their profile URL or @handle (YouTube, TikTok, Instagram) pulls their ~30 most recent videos in, scored against that channel's own median, and follows the channel as a competitor → poll `get_outlier_channel_ingest` until `done` → `list_outliers` with `channels: [channel.id]` (add `duration_type: shorts` for TikTok / Instagram). Video links belong to `fetch_outlier`, not this tool. ## Security & limits - Read-only credentials cannot write, anywhere. - Every AI tool call is recorded in the user's audit log (Settings → AI Assistant Access → activity). - Rate limits: 120 MCP requests/min per token; 180 `create_post`/hour; 40 `upload_media`/hour; 30 `search_outliers`/hour; 60 `fetch_outlier`/hour; 30 `generate_outlier_breakdown`/hour; 10 `add_outlier_channel`/hour. HTTP 429 = back off. - Rotate the API key any time to revoke access instantly.