# Vellarin Hub

> The tool layer for AI agents. One MCP URL and one key give Claude Code, Cursor, Codex or your own agent the apps your team connected in Vellarin Agents, with approval before anything is written, hard caps and a record of every call, plus a shared memory, a run log with tokens and cost, and eval results. A product of Vellarin AI (Vellarin Agents, Vellarin BidStream, Vellarin Hub).

Vellarin Hub is not a marketplace or reseller of third-party paid APIs. A workspace connects its own accounts (for example Gmail, Slack, a calendar, a CRM) once; its agents reach them through Hub.

## How it works
- Auth: `Authorization: Bearer vhub_...` (an agent key created on the Developers page of the workspace; shown once).
- MCP endpoint (Streamable HTTP, JSON responses): https://hub.vellarin.ai/api/hub/mcp
- REST API: https://hub.vellarin.ai/api/hub/v1 (GET /tools, POST /tools/{name}/call, GET /approvals/{id}, GET /usage, /memory, /traces, /runs, /evals)
- Read actions run at once. Write actions answer "waiting for approval" with an approval id; the workspace owner approves in Vellarin; the agent polls `vellarin_approval_status` and calls the same action again; it runs exactly once.
- Over a cap the answer is HTTP 429 "upgrade or wait" with the reset time. No wallet, no top-ups, no overage charges.

## Memory, run log and eval results (same key, same endpoint)
- Shared memory: a key-value store for the workspace's agents, so state survives between runs. MCP tools `vellarin_memory_set`, `vellarin_memory_get`, `vellarin_memory_list`, `vellarin_memory_delete`; REST GET /memory, GET | PUT | DELETE /memory/{key}. Values are text up to 16000 characters, with an optional time to live. Not a secrets store.
- Run log: an agent writes steps, model calls (tokens, cost), tool calls and errors; Hub adds them up per run. MCP tool `vellarin_trace_log`; REST POST /traces, GET /traces?run=, GET /runs.
- Eval results: record a score (0 to 1) or pass/fail per case; Hub answers pass rate and average score per suite. MCP tool `vellarin_eval_record`; REST POST /evals, GET /evals.
- These never reach a third party: no approval, and they do not count as app calls. Kept 30 days (run log and eval results).
- Hub Free: 100 memory keys, 5000 run-log events and 1000 eval results a month. Hub Team: 5000, 100000 and 50000.

## Limits
- Hub Free: 1 agent key, 500 calls a month, USD 0.
- Hub Team: 10 agent keys, 5000 calls a month, USD 19 a month or USD 180 a year. 14 days free, no card needed to start.
- 60 calls a minute per key.

## Quickstart
- Claude Code: `claude mcp add --transport http vellarin https://hub.vellarin.ai/api/hub/mcp --header "Authorization: Bearer <key>"`
- Cursor (~/.cursor/mcp.json) and Codex (~/.codex/config.toml): see the docs.

## Apps (one page each, with real action names)
- [Gmail](https://hub.vellarin.ai/apps/gmail): 61 actions
- [Google Calendar](https://hub.vellarin.ai/apps/googlecalendar): 47 actions
- [Google Sheets](https://hub.vellarin.ai/apps/googlesheets): 50 actions
- [Google Drive](https://hub.vellarin.ai/apps/googledrive): 92 actions
- [Google Docs](https://hub.vellarin.ai/apps/googledocs): 46 actions
- [Outlook](https://hub.vellarin.ai/apps/outlook): 317 actions
- [Slack](https://hub.vellarin.ai/apps/slack): 165 actions
- [Notion](https://hub.vellarin.ai/apps/notion): 57 actions
- [HubSpot](https://hub.vellarin.ai/apps/hubspot): 354 actions
- [Salesforce](https://hub.vellarin.ai/apps/salesforce): 288 actions
- [Stripe](https://hub.vellarin.ai/apps/stripe): 428 actions
- [QuickBooks](https://hub.vellarin.ai/apps/quickbooks): 114 actions
- [GitHub](https://hub.vellarin.ai/apps/github): 883 actions
- [Jira](https://hub.vellarin.ai/apps/jira): 166 actions
- [Linear](https://hub.vellarin.ai/apps/linear): 47 actions
- [Asana](https://hub.vellarin.ai/apps/asana): 247 actions
- [Trello](https://hub.vellarin.ai/apps/trello): 329 actions
- [ClickUp](https://hub.vellarin.ai/apps/clickup): 163 actions
- [Monday](https://hub.vellarin.ai/apps/monday): 126 actions
- [Airtable](https://hub.vellarin.ai/apps/airtable): 26 actions
- [Zendesk](https://hub.vellarin.ai/apps/zendesk): 480 actions
- [Intercom](https://hub.vellarin.ai/apps/intercom): 138 actions
- [Mailchimp](https://hub.vellarin.ai/apps/mailchimp): 275 actions
- [Calendly](https://hub.vellarin.ai/apps/calendly): 53 actions
- [Typeform](https://hub.vellarin.ai/apps/typeform): 35 actions
- [Dropbox](https://hub.vellarin.ai/apps/dropbox): 175 actions
- [OneDrive](https://hub.vellarin.ai/apps/one_drive): 101 actions
- [Excel](https://hub.vellarin.ai/apps/excel): 125 actions
- [Microsoft Teams](https://hub.vellarin.ai/apps/microsoft_teams): 237 actions
- [Discord](https://hub.vellarin.ai/apps/discord): 23 actions
- [Zoom](https://hub.vellarin.ai/apps/zoom): 94 actions
- [Figma](https://hub.vellarin.ai/apps/figma): 52 actions
- [Supabase](https://hub.vellarin.ai/apps/supabase): 230 actions
- [LinkedIn](https://hub.vellarin.ai/apps/linkedin): 29 actions
- [YouTube](https://hub.vellarin.ai/apps/youtube): 66 actions
- [WhatsApp](https://hub.vellarin.ai/apps/whatsapp): 127 actions

## Pages
- [Vellarin Hub](https://hub.vellarin.ai/): what it does, quickstart, approvals, limits, questions
- [Apps](https://hub.vellarin.ai/apps): the apps an agent can use through Hub
- [OpenAPI description](https://hub.vellarin.ai/openapi.json): the REST API as OpenAPI 3.1
- [API reference](https://hub.vellarin.ai/docs): authentication, MCP endpoint, approvals, REST API, errors, limits, security
- [Vellarin Agents](https://agents.vellarin.ai/en): the product where a team connects its apps and runs its own agents