> ## Documentation Index > Fetch the complete documentation index at: https://developers.mihu.ai/llms.txt > Use this file to discover all available pages before exploring further. # Mihu MCP Server > Connect LLM clients to Mihu via the Model Context Protocol Use this guide to connect the Mihu MCP Server to your client and start using it safely. * **Endpoint:** `https://mcp.mihu.ai/mcp` * **Transport:** Streamable HTTP (MCP standard) * **Health:** [`https://mcp.mihu.ai/health`](https://mcp.mihu.ai/health) ## Before you start Make sure you have: * access to the target Mihu tenant (e.g. `your-tenant.mihu.ai`), * an MCP-compatible client (Claude Desktop, IDE agent, or custom MCP client), * and a Mihu API token from your tenant panel. ## Access model Each token is tied to **one Mihu tenant**. The token inherits the panel permissions of the API key it was minted from — your client cannot exceed them. The MCP server itself stores no credentials. | Property | Value | | ---------------------- | -------------------------------------------- | | Tenant binding | One token, one tenant subdomain | | Permissions | Inherited from the underlying Mihu API token | | OAuth access token TTL | 1 hour (re-mint as needed) | | Storage | In-memory on the server (not persisted) | ## Connect Mihu MCP Server There are two ways to authenticate. Pick whichever your client supports. ### Option A — OAuth 2.1 (recommended for Claude Desktop) ```text theme={null} Authorization Server : https://mcp.mihu.ai Token URL : https://mcp.mihu.ai/oauth/token Grant types : client_credentials, authorization_code (PKCE) client_id : client_secret : ``` Claude Desktop discovers the OAuth flow automatically — no manual register step. In **Settings → Connectors → Add custom connector**: | Field | Value | | ------------------- | ------------------------- | | URL | `https://mcp.mihu.ai/mcp` | | OAuth Client ID | your tenant subdomain | | OAuth Client Secret | your Mihu API token | ### Option B — Direct headers (for scripts / curl / custom clients) ```text theme={null} Authorization : Bearer X-Mihu-Tenant : ``` No token mint needed; tenant is set per request. ### Mint a token directly ```bash theme={null} curl -s -X POST https://mcp.mihu.ai/oauth/token \ -d grant_type=client_credentials \ -d client_id= \ -d client_secret= # → {"access_token":"...","token_type":"Bearer","expires_in":3600} ``` Then call MCP with `Authorization: Bearer `. ## Test the connection Try a few read-only prompts first: * "List my agents." * "How many contacts do we have?" * "Show conversations from the last 24 hours." ## First setup checks After you connect, confirm these basics: * You are connected to the expected tenant. * Read tools (`list_agents`, `list_contacts`) return data. * Your panel permissions allow the actions you plan to run. ## What it can do The server exposes 375 tools across 43 modules. Common capabilities: | Capability | Example tools | | ------------------- | --------------------------------------------------------------------------- | | Place real calls | `initiate_call`, `say_in_call`, `forward_call`, `mute_agent`, `hangup_call` | | Send SMS / WhatsApp | `send_sms`, `send_whatsapp_template`, `whatsapp_call` | | Manage agents | `list_agents`, `create_agent`, `add_agent_guideline`, `add_agent_intent` | | Bulk outbound | `create_listing`, `add_contacts_to_listing`, `run_listing`, `stop_listing` | | Contacts | `list_contacts`, `create_contact`, `update_contact`, `add_tag_to_contact` | | Appointments | `book_appointment`, `list_appointments`, `get_available_slots` | | Conversations | `list_conversations`, `get_conversation`, `get_session_evaluation` | | Phone numbers | `search_available_phone_numbers`, `provision_call_channel` | Each tool's description includes required fields, enums, and example shapes — readable by the LLM at call-time, no external docs needed. ## What to include in a prompt Include these details when relevant: * **Time range** — "last 7 days", "yesterday", "this month" * **Target** — an agent, campaign, contact, conversation, or schedule * **Output** — summary, ranking, comparison, trend, or draft * **Action boundary** — "prepare only", "show preview", or "do not send" ### Example prompt patterns **Analysis** * "Summarize agent performance for the last 7 days." * "Which campaigns drove the most completed calls last week?" * "Compare SMS vs WhatsApp delivery rates for the last 14 days." **Audience and contact investigation** * "Show recent conversations for contact +90555…" * "List appointments booked in the last 3 days." * "Which contacts have the `vip` tag?" **Operational requests** * "Initiate a call from agent X to +90555… with greeting 'Hi…'" * "Create a listing for these 50 leads under agent Y." * "Stop the active listing named 'Q2 outbound'." ## How write actions work Write actions follow the underlying Mihu API permissions of the token: * The token inherits exactly what your API key can do in the panel — nothing more. * Bulk operations (e.g. `create_listing` with thousands of contacts) are accepted, but the agent's channel binding and tenant rate limits still apply. * All actions are logged in the standard Mihu request logs for the tenant. ## Troubleshooting | Symptom | First check | | ------------------------------ | ------------------------------------------------------------------------ | | `401` from upstream | Wrong tenant or wrong/expired API token. | | `404` from upstream | Tenant subdomain typo — does `.mihu.ai` resolve? | | "Missing tenant context" error | Client didn't send `X-Mihu-Tenant`, or OAuth token expired (1-hour TTL). | | OAuth token expired | Re-mint at `/oauth/token`. Default TTL is 1 hour. | | Action blocked by permissions | Underlying Mihu API key lacks the permission — adjust in the panel. | | Cannot access another tenant | Each token is tied to one tenant. Mint a separate token per tenant. | ## Support * **Status / health:** [`https://mcp.mihu.ai/health`](https://mcp.mihu.ai/health) → `{"ok": true}` * **Issues:** email [support@mihu.ai](mailto:support@mihu.ai)