MCP Integration - Share Of Model

Share Of Model exposes a Model Context Protocol surface so LLM agents — Claude.ai, Claude Desktop, Claude Code, Microsoft Copilot Studio, or any MCP-compatible client — can query your workspace data directly and set up new analyses with you.

What is exposed

Three things reach your client through the single MCP endpoint:

Tools — every GET endpoint of the API Reference, from both the Core API and the Search API, except account administration (API keys, platform integrations and accounts, user roles, favourites, SSO and SCIM). On top of these reads come the write tools below, get_current_user, which returns your roles so the assistant knows what you can do, and dashboard tools that render results as interactive views in clients that support MCP Apps.
Prompts — thirteen guided workflows you invoke by name. See Prompts below.
Skills — curated domain knowledge (metric glossaries, reporting playbooks, security and compliance) and step-by-step procedures to create brand analyses, Search analyses and GEO audits, which the assistant loads on its own, so you don’t have to explain what a metric means before asking about it. The documentation tools are also mounted, letting the assistant look things up in these docs while it works and report a problem with a page.

Write tools

Besides reads, the MCP exposes the write endpoints marked for agents:

Module	Tools
Brand Analysis	`create_analysis`, priced by `compute_analysis_cost`
Search Visibility	`create_search_visibility_analysis`, priced by `estimate_collection_cost`
GEO Audit	`create_website_evaluation`, priced by `compute_website_evaluation_cost`
Personas	`create_persona`, and `generate_personas` to have AI draft them
Asset Evaluations	`create_asset_evaluation`, priced by `compute_asset_evaluation_cost`, then `retry_asset_evaluation`, `archive_asset_evaluation` and `restore_asset_evaluation`. In the asset library, `bulk_edit_assets` renames and re-tags assets, `archive_assets` and `restore_assets` hide and bring them back.

create_asset_evaluation evaluates assets already in your library, so upload them in the web app first. Updating or deleting an analysis, and managing members, roles and API keys, are not exposed through the MCP.

Required role

Write tools act with your own Share Of Model role. You can create in a workspace when you are Organization Editor or Organization Owner, or Workspace Editor in that workspace — see User Roles & Workspaces. With a Viewer or Guest role only, write tools are refused; the assistant checks your role with get_current_user first and tells you.

Confirmation and cost

Before creating a brand analysis, a Search analysis or a GEO audit, the assistant follows the module’s creation procedure: it shows you any existing analysis that already covers your request, prices the exact setup with the matching cost tool, and shows you the full setup with its cost. It calls the create tool only after your explicit confirmation, and only once. It never creates or generates a persona without asking you first.
Write tools are annotated as not read-only, and the archive tools as destructive. MCP clients rely on these annotations to decide when to ask you before running a tool, so check your client’s tool permission settings and keep approval on for write tools.
Creating consumes credits or quota exactly as it does in the web app — see Credit System.

Server URL

https://mcp.shareofmodel.ai/mcp

Transport: Streamable HTTP.

Authentication

The server uses OAuth 2.0. You do not pass an API key or a JWT — on first connection, the client opens your browser, you sign in with your Share Of Model account (same credentials as the web app), and the client stores the resulting token. There is nothing to configure manually. The server publishes its OAuth 2.0 metadata at /.well-known/oauth-authorization-server and supports Dynamic Client Registration, so clients that rely on discovery — Microsoft Copilot Studio among them — register themselves without a pre-provisioned client ID or secret.

Connect from Claude Code

claude mcp add --transport http share-of-model https://mcp.shareofmodel.ai/mcp

The first time you invoke a tool, Claude Code opens your browser to complete the OAuth flow. Subsequent calls reuse the stored token.

Connect from Claude.ai or Claude Desktop

Open Settings → Connectors.
Scroll to the bottom and click Add custom connector.
Paste https://mcp.shareofmodel.ai/mcp and click Connect.
Sign in with your Share Of Model account in the browser window that opens.

The Share Of Model tools then appear in the tools panel of any new conversation.

Connect from Microsoft Copilot Studio

In your agent, open Tools → Add a tool → New tool → Model Context Protocol, paste the server URL, and select OAuth 2.0 → Dynamic discovery. Generative orchestration must be turned on. The full walkthrough, including publishing the agent to your whole organisation, is in Microsoft Copilot Studio.

Connect from MCP Inspector

npx @modelcontextprotocol/inspector

In the Inspector UI, choose Streamable HTTP as the transport, paste the server URL, then click Connect. You will be prompted to sign in on first use.

Prompts

Rather than composing a request yourself, you can invoke a pre-built workflow by name:

Prompt	What it produces
`brand_visibility_brief`	Executive brief on a brand’s AI visibility.
`compare_brands`	Head-to-head brand comparison.
`weekly_brand_update`	Week-over-week metrics summary.
`competitive_landscape`	Top brands in a category.
`persona_breakdown`	Performance by target persona.
`attribute_deep_dive`	Deep analysis of a single attribute.
`search_visibility_scorecard`	Search visibility scorecard.
`visibility_gap_vs_competitor`	Search query gaps versus a competitor.
`asset_effectiveness_report`	Creative asset evaluation summary.
`geo_opportunities`	Cross-module GEO recommendations.
`security_and_compliance_brief`	Security and compliance brief for an IT or procurement audience.
`setup_brand_analysis`	Sets up a brand analysis with you, then creates it once you validate the setup and its cost.
`setup_search_analysis`	Sets up a Search analysis for a website, then creates it once you validate the setup and its cost.

How you trigger a prompt depends on the client — in Claude Desktop and Claude Code they appear in the prompt or slash-command picker once the connector is authenticated. For the full walkthrough and example questions, see Get started with Share of Model MCP.