Remote MCP

The remote MCP lets you use SIPPulse AI from inside an MCP client: Claude, Cursor, VS Code, ChatGPT or any client that speaks Streamable HTTP with OAuth discovery. You authorize in the browser, choose the organization and the project, and the client gets the platform tools.

Every connection uses the same URL:

https://api.sippulse.ai/mcp

This is the SIPPulse AI MCP server itself. To connect your agents to third-party MCP servers, see MCP Servers and MCP Connections.

How the connection works

The MCP client discovers the OAuth configuration by itself, from the MCP URL:

1. The client reaches the URL and receives the authorization server location.
2. The browser opens the SIPPulse AI consent screen.
3. The screen shows the application asking for access, the organization, the project and the permissions involved.
4. You approve or deny. When you approve, the client receives a token and can call the tools.
5. The token refreshes automatically while the access stays valid.
6. If the token expires or is revoked, the client asks for authorization again.

Every consent is bound to one organization and to a project you can access. The access combines your current role in the organization with the granted permissions. You cannot widen the access later: to grant more permissions, the authorization has to be repeated.

Configure your client

Claude Code

claude mcp add --transport http sippulse https://api.sippulse.ai/mcp

Cursor and VS Code

In the client's mcp.json file:

{
  "servers": {
    "sippulse": {
      "type": "http",
      "url": "https://api.sippulse.ai/mcp"
    }
  }
}

Claude.ai (custom connector)

In Claude.ai, add a custom MCP connector and paste https://api.sippulse.ai/mcp as the server address. Other clients that accept a custom connector, such as ChatGPT, use the same URL and the same browser authorization flow.

API key for automation

For automations that do not go through interactive consent, use an API key in the Authorization header:

Authorization: Bearer <your_api_key>

An API key bound to a project fixes that project and does not prompt in the browser. It grants the full MCP tool catalog, so store it like any other credential. The key must have the same permissions the tool needs: the platform checks the key's RBAC, and an API key does not bypass permissions.

Permissions (scopes)

Scope	What it allows
spai:agents:read	Read agents and the internal tools catalog
spai:agents:write	Create, edit, validate and delete agents
spai:agents:execute	Run a real turn in a thread
spai:secrets:read	See secret names only
spai:channels:read	Read SIP, WhatsApp and Telegram deployments
spai:channels:write	Create, edit, activate, deactivate and delete deployments
spai:threads:read	Read thread metadata and a bounded history slice
spai:threads:write	Create, close and delete threads
spai:models:read	Read the organization model catalog

Reading, writing and executing are different permissions: reading never changes anything, writing creates and modifies, and executing actually runs a turn. A read-only authorization changes nothing, not even for owners and admins.

Available tools

Every tool is exposed by the MCP itself. The Cost column says whether the call consumes credits from your organization.

Agents

Tool	What it does	Scope	Cost
agents_list	Lists the agents in the effective project	spai:agents:read	No
agents_get	Returns one sanitized agent	spai:agents:read	No
agents_validate	Validates the model configuration without writing the agent	spai:agents:write	No
agents_create	Creates an agent without accepting secret material	spai:agents:write	No
agents_update	Updates an agent without accepting secret material	spai:agents:write	No
agents_delete	Deletes an agent after explicit confirmation	spai:agents:write	No
internal_tools_list	Lists the catalog of internal agent tools	spai:agents:read	No

Models

Tool	What it does	Scope	Cost
models_list	Lists the organization models and their capabilities	spai:models:read	No
models_get	Returns a model and its agent parameter schema	spai:models:read	No

Secrets

Tool	What it does	Scope	Cost
secrets_list	Lists secret names only; values and metadata are never returned	spai:secrets:read	No

SIP

Tool	What it does	Scope	Cost
sip_deployments_list	Lists SIP deployments	spai:channels:read	No
sip_deployments_get	Returns one SIP deployment	spai:channels:read	No
sip_deployments_apply	Applies a SIP deployment using project secret references	spai:channels:write	No

WhatsApp

Tool	What it does	Scope	Cost
whatsapp_deployments_list	Lists WhatsApp deployments	spai:channels:read	No
whatsapp_deployments_get	Returns one WhatsApp deployment	spai:channels:read	No
whatsapp_deployments_create	Creates a WhatsApp deployment	spai:channels:write	No
whatsapp_deployments_update	Updates a WhatsApp deployment	spai:channels:write	No
whatsapp_deployments_activate	Activates a WhatsApp deployment	spai:channels:write	No
whatsapp_deployments_deactivate	Deactivates a WhatsApp deployment	spai:channels:write	No
whatsapp_deployments_delete	Deletes a WhatsApp deployment after explicit confirmation	spai:channels:write	No

Telegram

Tool	What it does	Scope	Cost
telegram_deployments_list	Lists Telegram deployments	spai:channels:read	No
telegram_deployments_get	Returns one Telegram deployment	spai:channels:read	No
telegram_deployments_create	Creates a Telegram deployment using secret names	spai:channels:write	No
telegram_deployments_update	Updates a Telegram deployment	spai:channels:write	No
telegram_deployments_activate	Activates a Telegram deployment	spai:channels:write	No
telegram_deployments_deactivate	Deactivates a Telegram deployment	spai:channels:write	No
telegram_deployments_delete	Deletes a Telegram deployment after explicit confirmation	spai:channels:write	No

Threads

Tool	What it does	Scope	Cost
threads_list	Lists thread metadata without history	spai:threads:read	No
threads_get	Returns a bounded sanitized thread	spai:threads:read	No
threads_create	Creates a thread for an agent	spai:threads:write	No
threads_run	Runs one turn in a thread	spai:agents:execute	Yes
threads_close	Closes a thread	spai:threads:write	No
threads_delete	Deletes a thread after explicit confirmation	spai:threads:write	No

threads_run is the only tool that runs a real turn and is billed as a normal agent execution. agents_validate only checks the configuration, with no paid inference.

What the remote MCP never does

• It never returns secret values. secrets_list returns names only.
• It never deletes without confirmation: agents_delete and threads_delete require confirm: true in the request.
• threads_list never carries history, only metadata. threads_get returns a bounded, sanitized slice of the conversation.
• It does not expose audio transcription (STT).
• It has no streaming tool: every call is a JSON request.

Common errors

Error	What it means
401	The session expired or was revoked. Authorize the client again.
403 insufficient_scope	The authorization lacks the requested permission. Authorize again including that permission.

Never share access tokens or API keys. Whoever holds the token uses the MCP with your permissions. If a token leaks, revoke the access immediately.

Revoking access

You revoke any authorization at any time in Settings > OAuth access. See OAuth access. Revocation is immediate: from the next request on, the client has to authorize again.

Protocol note

The remote MCP uses the MCP 2026-07-28 protocol in stateless mode, with no session affinity between requests.