Back to Investor Brief

Investor brief · MCP server

Pause as a tool surface for every AI agent

Pause-Health.ai exposes its MuleSoft Experience APIs through a Model Context Protocol (MCP) server. Claude Desktop, Cursor, the Salesforce Agentforce Service Agent, and any MCP-compliant client can call Pause's clinical APIs as native tools.

The four tools

get_patient_timeline

Production: pause-patient-bundle-process-api

Returns a FHIR R5 Bundle (Patient + raw wearable Observations + DBDP-computed feature Observation with derivedFrom provenance).

get_patient_intake

Production: pause-intake-process-api

Returns the structured intake record produced by the Salesforce Agentforce Service Agent — chief complaint, symptom cluster, red-flag screen, triage recommendation.

find_menopause_providers

Production: pause-provider-directory-experience-api

Searches Pause's provider directory by ZIP and menopause-certified flag. Returns providers ranked by Pause's internal graph score.

experience_api_health

Production: pause-patient-bundle-process-api (root)

Liveness check for the Experience API plane. Use before larger tool calls.

Why MCP

One server, every modern agent

Claude Desktop, Cursor, OpenAI Responses API, and the Salesforce Agentforce Service Agent all speak MCP. Exposing Pause's MuleSoft Experience APIs as MCP tools makes them callable from every relevant AI runtime — no per-client adapter, no custom plugin format.

Production-grade contract today

The tool surface is identical between the mocked Experience APIs (today) and a customer's MuleSoft Anypoint deployment (tomorrow). Switching is a single environment variable: PAUSE_MCP_BASE_URL. Every agent integration built against the mock works against production with no client changes.

Layers cleanly on top of MuleSoft

MCP is the agent-side contract. MuleSoft is the integration-side contract. They compose: Mule Experience APIs become MCP tools 1:1, and Mule's API policies (rate limit, OAuth, observability) apply transparently to every MCP call.

Auditable provenance

Each tool call returns a meta block including which Process / Experience API would serve it in production, plus FHIR derivedFrom references for computed features. The same audit trail that satisfies HIPAA-compliant logging is visible to the agent at call time.

Touch the architecture

The MCP server lives at mcp/ in this repo. It speaks stdio (the standard transport for local AI clients) and fronts the mocked Experience APIs under /api/mulesoft/*. The descriptor is published at /.well-known/mcp.json so any MCP-aware registry or gateway can discover the tool surface.

GET /.well-known/mcp.jsonMocked Experience APITry the provider directoryMCP server source on GitHub

Register Pause in your MCP client

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "pause-health": {
      "command": "npx",
      "args": ["-y", "@pause-health/mcp"],
      "env": {
        "PAUSE_MCP_BASE_URL": "https://pause-health.ai"
      }
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json (or the project-local .cursor/mcp.json):

{
  "mcpServers": {
    "pause-health": {
      "command": "npx",
      "args": ["-y", "@pause-health/mcp"],
      "env": {
        "PAUSE_MCP_BASE_URL": "https://pause-health.ai"
      }
    }
  }
}

Salesforce Agentforce Service Agent

Production deployments register the Pause MCP server behind the customer's MuleSoft Anypoint gateway and expose it to Agentforce via the External Services connector. The tool surface is identical to local development.

// In Anypoint Platform / Agentforce Studio:
// 1. Publish the Pause MCP server behind an External Services connector
//    OR the Agentforce MCP gateway.
// 2. Point PAUSE_MCP_BASE_URL at the customer's Anypoint Experience-tier
//    base URL (set in the gateway / connector configuration).
// 3. The Agentforce Service Agent now sees four tools:
//    - get_patient_timeline
//    - get_patient_intake
//    - find_menopause_providers
//    - experience_api_health
// 4. Tool calls flow: Agentforce -> MCP gateway -> Pause MCP server
//    -> MuleSoft Experience API -> JupyterHealth / DBDP.

Prototype vs production

AspectPrototype todayCustomer deployment
MCP server transportstdio. Launched by the MCP client as a child process via npx.stdio for developer tooling (Cursor, Claude Desktop) and Streamable HTTP behind the customer's identity provider for Agentforce / server-to-server.
Backing Experience APIsMocked endpoints under /api/mulesoft/* served by the Next.js frontend (deterministic fixtures).Real MuleSoft Anypoint Experience APIs on the customer's Runtime Fabric or CloudHub 2.0. Same JSON shapes.
AuthenticationNone. Public read of synthetic demo data.Bearer token via PAUSE_MCP_API_KEY, validated by Mule API policies and tied to the customer's OAuth provider.
Tool surfaceFour tools: timeline, intake, providers, health.Same four tools; additional tools added as new Experience APIs ship (e.g. orders, referrals).
ProvenanceMocked meta blocks reference the production Mule API names.Real Mule correlation IDs, HIPAA audit IDs, and API instance identifiers returned in the meta block.

Phased plan

Phase 0 — Reference implementation

Today

Pause MCP server committed under mcp/. Four tools backed by mocked Experience APIs. Descriptor published at /.well-known/mcp.json. Investor page (this one).

Phase 1 — Publish to the registry

2 weeks

Publish @pause-health/mcp to npm. Submit to the public MCP server registry. Add a Streamable HTTP transport alongside stdio for server-to-server callers.

Phase 2 — Customer-managed gateway

4–6 weeks with first customer

Deploy the MCP server behind the customer's MuleSoft Anypoint API gateway. Wire OAuth via PingFederate / Azure AD. The Agentforce Service Agent registers it as a tool source.

Phase 3 — Multi-tenant MCP plane

Ongoing

One Pause MCP package, N customer deployments. Tools auto-expand as new MuleSoft Experience APIs ship. Telemetry rolled up across deployments for product analytics.

Why investors should care

Read deeper