The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Cavyro listing page.
Cavyro is a Telegram-native CRM for agents and small sales teams: contacts, companies, deals, pipelines, campaigns, docs and reports. This repository describes Cavyro's hosted Model Context Protocol server, listed in the MCP Registry as com.cavyro/cavyro.
The server is hosted, so there is nothing to install or run. Point your MCP client at it and it reads and acts on your Cavyro workspace with your permissions.
https://ai.cavyro.com/mcp (Streamable HTTP)https://ai.cavyro.com/.well-known/mcp/server-card.json (full tool surface with input schemas, no credentials needed)The server source is not published here. The tool surface is generated from the running server and exposed in the server card above, so the card is always the authoritative description of what the server does.
The server exposes a fixed set of nine generic tools over a resource and action registry, instead of one tool per API operation. The client discovers resources and actions on demand, so the tool list stays small while covering around 90 CRM operations.
| Tool | Purpose |
|---|---|
describe_resource | Catalog of all resources and global actions, or one resource's filters, sort fields, enums, operations and actions |
query_resources | List records of any resource with filters, sort, pagination and custom-field filters |
get_resource | Fetch one record by ID |
create_resource | Create a record (custom fields nested under custom_fields) |
update_resource | Change fields on a record |
delete_resource | discard (archive), restore, or destroy a record |
search | Omnibox search across entity types |
perform_action | Any specialized non-CRUD action by slug: move a deal between stages, link records, bulk operations, comments, reports |
load_guidance | On-demand workflow guidance: create flows, custom fields, automations |
Covered domains: deals, contacts, companies, pipelines and stages, comments and reactions, activity feed, notifications, workspaces, members, teams, groups, invitations, custom field definitions, automations, docs with their pages, reports, and read-only Telegram chats.
All tool responses share one envelope, { "data": ..., "pagination": ... }, with pagination present only on paginated list results.
The connection acts as you. An external AI can only see and do what your Cavyro role allows in that workspace.
Every request needs two headers:
| Header | Value |
|---|---|
Authorization | Bearer <cavyro_api_token> |
X-Workspace-Id | Your workspace ID |
Create an API token in Cavyro under Settings → API tokens. Tokens can expire after 7, 30 or 90 days, or never for integrations that cannot rotate them. Revoking a token cuts the client off immediately.
There is no OAuth flow. The server answers unauthenticated requests with 401 and a WWW-Authenticate: Bearer realm="cavyro" challenge. MCP is a Pro feature; MCP calls do not consume Cavyro AI credits, the model tokens are paid by your own client subscription.
Replace <your-api-token> and <your-workspace-id> in the snippets below. Cavyro's settings page generates the same snippet with your workspace ID filled in.
.vscode/mcp.json, or MCP: Add Server from the command palette:
Cursor: ~/.cursor/mcp.json. Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json.
Add the server as a connector in Developer Mode (Plus, Pro, Business or Enterprise) with the URL https://ai.cavyro.com/mcp and the two headers above.
Once connected, ask your assistant in plain language. A few things that work well:
The assistant will typically call describe_resource first to learn the shape of a resource, then query_resources or perform_action. If it needs a multi-step recipe, it calls load_guidance.