MCP Provider
Feature Overview
PPP exposes its data (CPD, Logbook, Profile, Appraisals, Wellness — scoped to the authenticated doctor) as a Model Context Protocol (MCP) provider. Doctors connect their own AI agents (Claude, OpenAI, Gemini) and query their portfolio without using the PPP UI — supporting the "manage records on your own device" goal.
Open Questions
| # | Question | Proposed answer / status | Ref |
|---|---|---|---|
| Q1 | OAuth scopes — what's the minimum read-only set; what can be written (CPD entry, reflection)? | Confirmed (transcript 2026-05-20 ~00:43:23): Exposing platform as MCP provider is confirmed. Matt agreed immediately. Scope/rate-limit/token details still TBD — see remaining rows below. Engineering proposal: v1 read-only across CPD / Logbook / Profile; CPD-entry write in v2 after audit. | Transcript 2026-05-20 §00:43:23 |
| Q2 | Rate limiting and abuse — per-doctor request budget? | Engineering proposal: 100 req/min, 5000/day per token. | Engineering-raised |
| Q3 | Admin exposure — do admin tiers get their own MCP scope for cohort queries? | Engineering proposal: no (v1) — admin queries go via dashboard; revisit if Matt asks. | Engineering-raised |
| Q4 | Token lifecycle — fixed expiry vs refresh-token? | Engineering proposal: 90-day expiry, doctor-revocable from profile settings. | Engineering-raised |
Closed Questions (0) — click to expand
| # | Question | Resolution | Ref |
|---|
Matt's Requirements (from Notion brief)
Source: transcript only — no Notion page yet authored.
Purpose
- Expose PPP as a Model Context Protocol (MCP) provider so doctors interact with their CPD / portfolio data via Claude, OpenAI or Gemini — without using the PPP UI.†
- Aligns with the goal of doctors managing records on their own devices.†
Surface (v1)
- Read tools for: CPD activities, Surgical Logbook entries, Profile (registrations, hospitals, qualifications), Wellness self-reports.
- Write tools (v1 minimum): the doctor's AI must be able to "grab this email from my inbox and update my CPD" and "ask for a report for my appraisal" — i.e. ingest-from-inbox into CPD and request-report flows.†
- Demo priority: explicitly de-prioritised vs the 28-May demo — ship after demo if scope tightens.†
Acceptance Criteria
Core — MCP server
- ⏳ PPP exposes an MCP server at a public endpoint (e.g.
mcp.professionalperformanceportfolio.com).† - ⏳ Server implements the published MCP spec (stdio + HTTP transports).
- ⏳ Discovery endpoint lists available tools and resources for an authenticated doctor.
Authentication
- ⏳ Doctor generates a personal-access MCP token from Profile → Integrations.
- ⏳ Token is HMAC-SHA256 signed; carries the doctor's userId + scope claims.
- ⏳ Token expiry default 90 days, doctor-revocable any time (Q4).
- ⏳ Token never exposes admin/cohort data — only the issuing doctor's records.
v1 read tools
- ⏳
list_cpd_activities(year?, category?)— returns CPD activities for the doctor. - ⏳
get_cpd_compliance_summary()— returns annual progress (hours, categories, AMC requirements). - ⏳
list_logbook_entries(date_from?, date_to?, procedure?)— returns surgical logbook entries. - ⏳
get_profile()— returns Profile (registrations, hospitals, qualifications). - ⏳
list_wellness_responses(date_from?)— returns doctor's own PPP Pulse / WHO-5 responses.
v1 write tools
- ⏳
add_cpd_activity(activity)— create a CPD activity from a parsed email / external source.† - ⏳
request_appraisal_report(timeframe)— async generation of an appraisal PDF, returned as a download URL.† - ⏳ Write tools require step-up confirmation (signed callback) so an AI cannot mutate state without doctor acknowledgement.
- ⏳ Every write tool invocation appears in the doctor's own activity history with the source-AI identifier captured.
Rate limiting
- ⏳ 100 req/min, 5000/day per token (Q2). 429 on overage.
- ⏳ Server emits structured rate-limit headers.
Audit & telemetry
- ⏳ Every MCP tool invocation is audit-logged (token id, tool name, timestamp, response size).
- ⏳ Doctor can view their own MCP access log from Profile → Integrations.
- ⏳ Anomalous access (e.g. 10× normal volume) triggers a security email to the doctor.
Documentation
- ⏳ Public docs page lists available tools + example payloads + how to wire to Claude/Codex.
- ⏳ Docs include a one-paragraph data-policy summary (read-only, doctor-scoped).
Current Status
| Sub-feature | Status | Notes |
|---|---|---|
| MCP server scaffold | ❌ Not started | |
| Token issuance + revocation | ❌ Not started | |
| v1 read tools | ❌ Not started | |
| Rate limiting | ❌ Not started | |
| Audit log + doctor-visible access UI | ❌ Not started | |
| Public docs page | ❌ Not started |
Open issues on label:feature:mcp-provider — no open issues. Fetched at: 2026-10-13T07:29:49Z — live list on GitHub.
Downstream References
| Feature | Linkage |
|---|---|
| Profile (Doctor) | Source of profile data |
| CPD | Read tools surface CPD activities + compliance summary |
| Logbook | Read tools surface logbook entries |
| Wellness | Read tools surface PPP Pulse responses |
| API Licensing | Sibling spec — paid B2B API; MCP is doctor-personal |
Change Log
| Date | Change | PR/Issue | Reason |
|---|---|---|---|
| 2026-07-13 | BDD scenario coverage intentionally deferred — spec is status: specified (not yet built). No @feature:mcp-provider scenarios needed yet. | #10578, #10572 | Coverage/BDD gap audit (#10572) flagged this spec as having zero BDD tag match. Confirmed specified-but-unbuilt per front-matter status:; BDD is written when the feature is built, not at the specification stage. This note stops future coverage sweeps from re-flagging the gap. |
| 2026-05-20 | New spec — expose PPP as Model Context Protocol (MCP) provider; doctor-scoped read tools for CPD, Logbook, Profile, Wellness; doctor-issued tokens. | #0 | Matt's 2026-05-20 Dev Session ratified exposing PPP as an MCP provider so doctors can interact with their portfolio via Claude/OpenAI/Gemini without using the UI. |