CIPP MCP Server

MCP (Model Context Protocol) server for CIPP — the CyberDrain Improved Partner Portal. Provides AI assistants with structured access to CIPP's M365 multi-tenant management capabilities.

Features

• 49 tools across 12 categories
• Tenant, user, group, and mailbox management
• Mailbox and online-archive size reporting, per tenant or per user
• Per-user Entra ID sign-in logs (status, location, Conditional Access, MFA)
• Security: Conditional Access policies, named locations, Microsoft Secure Score
• Standards & compliance: BPA, domain health, drift detection
• License reporting (per-tenant and CSP-wide)
• Alerts, audit logs, and scheduled tasks
• GDAP role and invite management
• Stdio and HTTP transport modes
• MCP Gateway compatible

Prerequisites

• Node.js 18+
• A running CIPP deployment
• CIPP API Key (generated from CIPP Settings → API Client Management)

Installation

Via npm (once published)

npx cipp-mcp

From source

git clone https://github.com/WYRE-AI/cipp-mcp
cd cipp-mcp
npm install
npm run build

Configuration

Set these environment variables (or copy .env.example to .env):

/health stays unauthenticated and does not read credentials. The stdio transport does not use CONDUIT_S2S_SECRET.

HTTP authentication

With MCP_TRANSPORT=http and CONDUIT_S2S_SECRET set, every /mcp request must include X-Gateway-S2S. The header value is t=<unix seconds>,v1=<hex>, where v1 is the hex-encoded HMAC-SHA256 of the ASCII string t=<unix seconds> keyed with CONDUIT_S2S_SECRET. The timestamp must be within 300 seconds of the server clock. A missing or invalid header is HTTP 401, before CIPP credentials are read. The secret is never written to logs.

X-Gateway-S2S: t=1710000000,v1=<64 lowercase hex chars>

In AUTH_MODE=gateway, /mcp also requires x-base-url plus either x-api-key or all of x-tenant-id, x-client-id, and x-client-secret (x-token-scope, x-token-url, and x-token-scope-fallback are optional). Those headers are the only CIPP credential source for that request. A request missing them is HTTP 401 and does not use CIPP_* from the process environment.

AUTH_MODE=env still requires a valid X-Gateway-S2S header, then uses CIPP_* from the environment.

Local HTTP without a secret is development-only. It binds 127.0.0.1 even when MCP_HTTP_HOST is 0.0.0.0 or another non-loopback address:

MCP_TRANSPORT=http MCP_ALLOW_INSECURE_DEV=1 npm run dev

Docker

The image listens on 0.0.0.0 inside the container (MCP_HTTP_HOST) and defaults to AUTH_MODE=gateway. Publish the port on loopback and pass the same S2S secret the gateway uses. In gateway mode, CIPP credentials arrive on each request as headers; they are not read from the environment.

docker build -t cipp-mcp .
docker run --rm \
  -e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \
  -p 127.0.0.1:8080:8080 \
  cipp-mcp

Compose refuses to start when CONDUIT_S2S_SECRET is unset. The cipp-mcp service is gateway mode: it sets AUTH_MODE=gateway, publishes 127.0.0.1:8080:8080, and does not set CIPP_API_KEY, CIPP_CLIENT_ID, CIPP_CLIENT_SECRET, CIPP_TENANT_ID, CIPP_BASE_URL, or env_file.

CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" docker compose up

Single-tenant HTTP (AUTH_MODE=env) still requires the S2S secret, then uses CIPP_*:

docker run --rm \
  -e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \
  -e AUTH_MODE=env \
  -e CIPP_BASE_URL=https://cippXXXXX.azurewebsites.net \
  -e CIPP_TENANT_ID=your-entra-tenant-id \
  -e CIPP_CLIENT_ID=your-client-id \
  -e CIPP_CLIENT_SECRET=your-client-secret \
  -p 127.0.0.1:8080:8080 \
  cipp-mcp

Usage with Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "cipp": {
      "command": "node",
      "args": ["/path/to/cipp-mcp/dist/entry.js"],
      "env": {
        "CIPP_BASE_URL": "https://cippXXXXX.azurewebsites.net",
        "CIPP_TENANT_ID": "your-entra-tenant-id",
        "CIPP_CLIENT_ID": "your-client-id",
        "CIPP_CLIENT_SECRET": "your-client-secret"
      }
    }
  }
}

Note: CIPP_BASE_URL must be the Azure Function App URL (.azurewebsites.net), not the frontend SWA URL (.azurestaticapps.net or your custom domain). The SWA enforces browser-based auth and will redirect all API requests to a Microsoft login page.

Tools

Editing a user right after creating it

edit_user first resolves the account's current UPN, because CIPP's Invoke-EditUser rebuilds the UPN from username + Domain on every call and editing without it would rename the account. Graph can lag a few seconds behind a just-created account, so when that lookup finds nothing, answers HTTP 404, or fails with a 5xx whose body is Graph's not-found (the Request_ResourceNotFound code, or Graph's "Resource '' does not exist or one of its queried reference-property objects are not present" message), it is retried up to 4 attempts in total, waiting 1s, 3s and 6s (~10s). The decision uses the HTTP status, not message text; any other 4xx (400, 401, 403) and other 5xx are not retried. If the user still isn't visible the error says "Tried N times" and, when the last attempt got an HTTP error, its status; wait a moment and retry.

Mailbox and archive sizes

list_mailbox_usage reports every mailbox in a tenant — primary size, item count, quota, percent of quota, and the same four figures for the online archive — sorted largest first, with tenant-wide totals that cover every mailbox even when only the top rows are returned.

It requires CIPP's reporting database to have been synced for that tenant. This is not a design choice: Invoke-ListMailboxes' live Exchange query selects no size fields at all, so the cache is the only tenant-wide source of sizes. Sync it from CIPP under Reports → Report Settings. If it has not been synced the tool says so and names the remedy rather than returning an empty result.

get_mailbox_usage reports the same figures for a single mailbox and reads live, so it needs no cache. Prefer it when the cache is unavailable or stale.

Two caveats worth knowing:

• Concealed report names blank the tenant-wide sizes. With Reports: conceal user, group, and site names enabled in the Microsoft 365 admin centre, Graph's usage report returns 32-character hashes instead of UPNs, so CIPP's join against the mailbox list matches nothing and every mailbox caches a size of 0 — a tenant that reads as empty rather than as failed. list_mailbox_usage detects this and returns a warning alongside the totals. get_mailbox_usage is unaffected: it reads the Exchange admin API directly.
• Sizes are gigabyte-rounded on the per-user path. CIPP rounds to two decimal places of a gigabyte before returning, so get_mailbox_usage byte counts are accurate to roughly 10 MB. Quotas are exact — they are recovered from the raw Get-Mailbox string, which carries the true byte count.

User sign-in logs

list_user_signin_logs returns one user's most recent interactive sign-ins, newest first: time, app, IP, location, success or failure with error code and reason, client app, Conditional Access status and the policies that evaluated, the authentication methods Graph recorded (first factor included — whether MFA was required is authenticationRequirement), device, and risk when flagged — plus a summary of failures, distinct IPs and countries.

• Accepts a UPN or an object id, and always resolves it first. CIPP filters Graph on userId, which is the Entra object id; a UPN there matches nothing and returns an empty page that reads as "this user never signs in". An unresolvable user is an error, not an empty result.
• One tenant, one user, one page. top defaults to 50 and caps at 1000; a full page carries a warning that older sign-ins may exist. allTenants is rejected — the endpoint has no all-tenants branch. Tenant-wide sign-ins are CIPP's Sign-Ins report (ListSignIns), which this server does not expose.
• Needs Entra ID P1/P2 in the tenant. Graph refuses sign-in log API access otherwise; the tool names the licence gap instead of relaying a generic "Failed to retrieve Sign In report".

Microsoft Secure Score

get_secure_score returns one tenant's latest Secure Score and its recent daily history. list_secure_score_controls returns the recommendation catalog: titles, rank, max points, cost, remediation, threats, and any decision to mark a control resolved by a third party or ignored.

CIPP has no dedicated Secure Score read endpoint. The only Secure Score HTTP function is ExecUpdateSecureScore, which updates a control and is not exposed here. Both tools call ListGraphRequest, which is what CIPP's own Secure Score page does, with Endpoint set to security/secureScores or security/secureScoreControlProfiles.

• One tenant. allTenants is rejected. CIPP's estate view is the nightly score cache, not a live Graph call, and these tools do not read that cache.
• top limits history only when pagination is turned off. ListGraphRequest treats $top as Graph's page size and follows every next link unless NoPagination is set. get_secure_score always sends both, so top (default 7, maximum 90) is the number of daily readings returned. The control catalog leaves pagination on, with a page size of 999, so the full list comes back.
• An empty result is not a score of zero. A tenant that has not been scored yet, or whose CIPP app is missing SecurityEvents.Read.All, comes back as no reading with a warning. A permission denial names that Graph permission instead of a generic CIPP HTTP 400.
• Comparisons are matched by basis. "All tenants" is AllTenants and "similar tenants" is TotalSeats. Those figures are already percentages. The percentage of the tenant's own score is currentScore / maxScore, and it is omitted when the maximum is missing rather than reported as 0%.

CIPP version compatibility

Request bodies are shaped against CIPP's own Invoke-*.ps1 handlers and are written to satisfy both current and older CIPP builds — where the two differ, the server sends the form both accept. Three behaviours are worth knowing:

• offboard_user reports queued, not completed. CIPP's ExecOffboardUser returns HTTP 200 the instant the job is created; it never waits for or reports the offboarding result. Confirm the outcome in CIPP's Offboarding view before treating an account as offboarded. The tool refuses a call with no actions selected, since that would otherwise queue a job that succeeds while doing nothing.
• Some endpoints report failure under HTTP 200. EditUser, AddScheduledItem and ExecOffboardUser return error text in Results rather than an error status. These tools parse Results and return status: "failed"; do not treat a 200 as success.
• Two parameters need a recent CIPP. offboard_user's DisableOneDriveSharing and set_out_of_office's timezone are ignored by older builds rather than erroring — so an offboarding that selects only DisableOneDriveSharing will run no actions on an older CIPP.

Authentication Setup

CIPP's API Client Management page provisions an Entra ID app registration and returns an OAuth client ID + client secret (not a long-lived Bearer token). The server exchanges these for a short-lived access token on each request using the OAuth 2.0 client-credentials flow, and caches the token until just before its expiry.

1. In CIPP, go to Settings → CIPP Settings → Integrations → CIPP-API
2. Create a new API client
3. Copy the Client ID and Client Secret — you will not be able to retrieve the secret later
4. Configure the server with the Function App URL (see below):
   envCopy

   CIPP_BASE_URL=https://cippXXXXX.azurewebsites.net
   CIPP_TENANT_ID=<your-entra-tenant-id>
   CIPP_CLIENT_ID=<client-id-from-cipp>
   CIPP_CLIENT_SECRET=<client-secret-from-cipp>
   

If you already have a static Bearer token (older CIPP deployments), set CIPP_API_KEY instead and leave the OAuth variables unset. When both are provided, CIPP_API_KEY wins.

The access token is requested for api://<clientId>/.default. That is the Application ID URI CIPP's App Service authentication allows. A token requested for the bare <clientId>/.default scope has its aud set to the client id GUID, and App Service auth rejects it with HTTP 401 and an empty body. Deployments that still expect that legacy audience are handled automatically: when no explicit scope is configured, a 401 is retried once with the other automatic audience (api://<clientId>/.default or <clientId>/.default). The audience that succeeds is reused for later calls from the same client. If that audience later starts failing with 401, the pin is dropped and the other audience is tried once, then whichever succeeds is remembered again. The retry does not run for any other status (403, 500, and so on), and a single request never tries more than once. Set CIPP_TOKEN_SCOPE to force a scope, or CIPP_TOKEN_SCOPE_FALLBACK=false (alias TOKEN_SCOPE_FALLBACK, gateway header x-token-scope-fallback) to skip the retry.

Finding your Function App URL

CIPP runs as an Azure Static Web App (SWA) backed by an Azure Function App. The SWA URL (your custom domain or *.azurestaticapps.net) enforces browser-only auth and cannot be used as CIPP_BASE_URL. Use the Function App URL instead.

Self-hosted CIPP: Find the Function App in the Azure portal (look for an App Service with Kind: functionapp in the same resource group as your SWA), or run:

az staticwebapp show --name <your-swa-name> --resource-group <rg> \
  --query "linkedBackends[0].backendResourceId" -o tsv

CIPP-sponsored hosting: Contact the CIPP team for your instance's Function App URL — it is not the same as the URL shown in your browser.

IP Allowlist

CIPP validates each API client against an IPRange field stored in Azure Table Storage. If your server's public IP is not in this list, you will receive:

Access to this CIPP API endpoint is not allowed, the API Client does not have the required permission

Self-hosted: Add your IP via the CIPP UI (Settings → API Client Management) or directly in the ApiClients table of your CIPP storage account.

CIPP-sponsored hosting: Ask the CIPP team to add your server's public IP to your API client's allowed range.

License

Apache-2.0 — see LICENSE

Contributing

Issues and PRs welcome. This server is tracked against wyre-technology/msp-claude-plugins#24.