covedataprotection-mcp

covedataprotection-mcp

MCP server for Cove Data Protection (N-able's backup/BDR platform,formerly N-able Backup / Backup Manager). Exposes the Backup ManagerJSON-RPC Management Service API as MCP tools.

MSPbots Connector gives your AI clients one MCP URL to reach your entire MSP stack (RMM, PSA, documentation, security, finance), and gives you control over who can call which tool and a record of everything that was called.

Hosted service: connect.mspbots.ai · Free to use.

Why Connector

Quick start

1. Sign up at connect.mspbots.ai.
2. Connect your vendors. Pick from 150+ connectors and authorize each one on the vendor's own page, or bring your own API keys. No extra vendor accounts needed.
3. Set access. Create roles and decide which tools each role may call. Nothing is allowed until you allow it.
4. Add your MCP URL to your AI client. Copy the URL from your Connector dashboard into Claude, ChatGPT, Cursor, VS Code, Gemini, Codex, or any MCP-capable client.
5. Ask your AI to do real work, and watch every call appear in the audit log.

Features

Unified gateway

• One MCP URL for all your AI clients
• 150+ connectors across the MSP stack
• Custom MCP servers supported, free
• Authorization happens on each vendor's own page; the AI never sees passwords or tokens
• Bring your own credentials: use the vendor API keys you already have
• Multiple accounts per app; connect once, reuse across AI clients
• Tool search: finds the right action for the task
• One sign-in; vendor credentials are injected behind the scenes and OAuth tokens refresh automatically

Visibility

• Audit log for every tool call (who, which vendor, when)
• Usage dashboard: call volume, top vendors, error rate, per-user activity
• Export audit logs to CSV, or stream to Loki, Graylog, or LogScale (SIEM)
• Optional recording of tool arguments and result summaries (off by default, enabled per organization)

Control

• Per-role, per-tool allowlist, deny by default
• owner / admin / member roles
• SCIM provisioning with Entra ID, Okta, JumpCloud, and Google Workspace

Security

• Credentials encrypted one record at a time (AES-256-GCM, per-record key)
• AI clients authenticate with OAuth 2.1 + PKCE
• Vendor responses are scanned for prompt-injection patterns and returned as structured data. This is one layer of defense, not a guarantee against every attack.
• Static egress IP for customer firewall and API allowlisting

Pricing

Free.

Links: connect.mspbots.ai · mspbots.ai

Overview

• Stateless HTTP service. No credentials are ever persisted — each requestsupplies its own credentials via headers, used only for the lifetime ofthat single request.
• Supports concurrent requests; per-request credential isolation is done viaPython contextvars, not a global/shared client instance.
• Entry points: POST /mcp (MCP protocol) and GET /health (health check).
• Default port: 8080 (configurable via MCP_HTTP_PORT).

Scope

26 tools, trimmed down from an original 247-tool full-schema build(2026-08-04). MSPbots' own Cove Data Protection integration (confirmed livevia /web/int/sys/integration/api/list against the production platform) isconfigured with exactly 6 APIs, mapped to these underlying JSON-RPC methods:

The other 5 confirmed-real methods above were kept as-is, plus same-categorycore CRUD (Add/Get/Modify/Remove) for each of the three resourcetypes they touch — accounts (9 tools), partners (9 tools), users (7tools) — plus GetServerInfo (1 tool, connectivity self-test, used for theoriginal live-verification below). Every other category from the original247-tool build (notifications, storage_nodes, contacts, storage,view_delivery, labels, jobs, custom_columns, branding, products,eula, locations, countries, audit, email, regions, features,permissions, templates — 19 categories, ~221 tools) was removed entirelyas unused by MSPbots and out of scope for this server's purpose.

Authentication

Cove has no static long-lived API key. Instead, every session starts with aLogin call (partner + username + password) that returns ashort-lived visa token (valid ~15 minutes), which must accompany everysubsequent call.

This server does not cache a visa across requests — caching one wouldmean persisting session state, which conflicts with the "no credentialpersistence" requirement. Instead, api_client.CoveClient.call() performs afresh Login on every tool invocation and discards the resulting visaafterward, trading one extra HTTP round trip per call for full statelessness.

HEADER authentication parameters

Missing any of the three headers returns 401:

{
  "error": "Missing credentials",
  "message": "This server requires the X-CoveDataProtection-Partner, X-CoveDataProtection-Username, and X-CoveDataProtection-Password headers",
  "required_headers": ["X-CoveDataProtection-Partner", "X-CoveDataProtection-Username", "X-CoveDataProtection-Password"],
  "optional_headers": []
}

Environment Variables

MCP Endpoint

• POST /mcp — MCP protocol (streamable HTTP transport)
• GET /health — health check, returns exactly {"status": "ok"} (pure local liveness probe; does not call the Cove API)

Tool List

Tool names follow covedataprotection_<snake_case_method_name> — e.g. theJSON-RPC method EnumerateAccountStatistics becomescovedataprotection_enumerate_account_statistics. Parameter names mirrorthe JSON-RPC method's own parameter names (camelCase → snake_case); complexstruct/enum-typed parameters are accepted as a dict/str and passedthrough to the vendor API as-is — see Known Gaps for what that means inpractice.

Test Example

# Health check
curl -s http://localhost:8080/health

# Call a tool via the MCP protocol (streamable HTTP) — requires an
# initialize handshake first per the MCP spec; abbreviated example below
# shows the tool-call request body only:
curl -s -X POST http://localhost:8080/mcp \
  -H "X-CoveDataProtection-Partner: <your-partner-name>" \
  -H "X-CoveDataProtection-Username: <your-username>" \
  -H "X-CoveDataProtection-Password: <your-password>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: <session-id-from-initialize>" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "covedataprotection_get_server_info",
      "arguments": {}
    }
  }'

Expected: 200 with the server version on valid credentials; on invalidcredentials, the Login call itself fails and every tool returns astructured JSON error envelope, e.g.:

{"error":{"code":"unauthorized","message":"<vendor error message>","retryable":false}}

code is one of the fixed values not_configured / unauthorized /not_found / invalid_argument / rate_limited / upstream_error;retryable tells the caller whether retrying the same call could succeed(true only for rate_limited and upstream_error). Tools never raiseexceptions for business/vendor errors — this envelope is returned as thenormal tool result string.

Live-verified (2026-07-29): covedataprotection_get_server_info (zeroparameters) was called end-to-end through this running server with a realpartner/username/password and returned the actual Cove server version —confirming the Login → visa → method-call pipeline works against the liveAPI, not just structurally.

API Reference

• Getting Started: https://developer.n-able.com/n-able-cove/docs/getting-started
• Authorization: https://developer.n-able.com/n-able-cove/docs/authorization
• Constructing a call: https://developer.n-able.com/n-able-cove/docs/construct-a-json-rpc-api-call
• Full method/struct/enum schema (JSON): https://documentation.n-able.com/covedataprotection/Schema_23.3.json

Known Gaps

• Trimmed from 247 to 26 tools on 2026-08-04. The original build coveredevery method in the Management Service schema. Per a later scope decision,this was cut down to what MSPbots' own production integration configactually calls (verified live via /web/int/sys/integration/api/listagainst app.mspbots.ai, integration id 2026570354981494786) plussame-category core CRUD — see the Scope section above for the exactAPI→method mapping and the full list of removed categories. The removed~221 tools (notifications, storage_nodes, contacts, storage,view_delivery, labels, jobs, custom_columns, branding,products, eula, locations, countries, audit, email, regions,features, permissions, templates) are not in this build at all; if afuture need requires one of them, the original schema(Schema_23.3.json, linked below) still documents its exact methodsignature and it can be re-added the same way the kept tools weregenerated.
• This covers the Management Service only. Cove's Getting Started guidedescribes a second, separate Reporting Service ({host}/repserv_json,runs per storage node) that provides backup/restore session statistics.MSPbots' own configuration calls a "Query Sessions" report against thisReporting Service — it is not part of the Management Service schemathis MCP is generated from, and no public schema for the ReportingService was found. This is the one MSPbots-configured endpoint this MCPdoes not cover; it can be added if the Reporting Service's methodlist/schema becomes available.
• Complex struct/enum parameters are untyped (dict/str) rather thanfully modeled. The schema defines 106 structs and 78 enums; fullyreproducing each one as a typed Python parameter was out of scope for amechanically-generated server. Callers need to shape these dictarguments to match the vendor's schema (see the Structs section ofSchema_23.3.json for exact field names) — each such parameter'sdescription names the struct it maps to and points at that schema file.
• No visa caching — see the Authentication section above. Every toolcall performs its own Login, which is simple and fully stateless butmeans 2 HTTP requests to the vendor per tool call instead of 1.
• All 26 tools were code-generated directly from the vendor's own schemafile, not hand-written — parameter names/types are only as accurate asthat schema. covedataprotection_get_server_info was the one toollive-verified end-to-end; the rest are structurally correct (schemavalidated, MCP-protocol tools/list confirmed) but not individuallysmoke-tested against real data.

Vendor MCP SOP Compliance Notes

• Tool count (26) exceeds the SOP's "should be ≤20" guideline. Alreadytrimmed 247 → 26 (see above); the remaining 26 map 1:1 to distinctJSON-RPC methods across 3 resource domains (accounts/partners/users) thatcannot be merged without losing required-parameter distinctions (e.g.get_*_info by name+password vs. get_*_info_by_id by numeric ID aredifferent lookup keys, not the same tool with an optional parameter).
• No limit / page_size scalar parameter exists on any tool. None ofthe 26 methods expose a first-class numeric page-size parameter — the onepagination-shaped field (range on EnumerateChildPartners) is an opaquevendor struct passed through as-is, and children_limit onGetPartnerTree is likewise passed straight to the vendor uninterpreted.There is nothing for this server to clamp against the SOP's default-50 /hard-cap-200 guidance; if the vendor schema's range/limit structs turnout to have their own documented per-page maximum, that should beenforced here instead of the SOP's generic fallback ceiling.
• No credential env-var fallback. config.Settings has nopartner/username/password field; get_client_from_context returns None(→ not_configured envelope) whenever the per-request contextvar isunset. There is no code path that falls back to an environment variable.
• Tool annotations: every read tool (get_*, enumerate_*) is markedreadOnlyHint=True, idempotentHint=True; add_* tools arereadOnlyHint=False, idempotentHint=False (creating twice is not ano-op); modify_*/set_*/remove_* tools arereadOnlyHint=False, destructiveHint=True, idempotentHint=True.