Give the AI real power — without handing over the keys.
Most AI connectors stop at writing posts, because everything beyond that is dangerous. AlphaBridge goes further — files, database, deployment — and treats that danger as an engineering problem: encrypted credentials, verified targets, allowlists, and a trace for every single action.
What makes the dangerous parts safe.
Content tools are easy. These three layers are what let AlphaBridge safely offer the tools nobody else offers.
Encrypted credentials
Deploy passwords are encrypted with libsodium (XSalsa20-Poly1305) before they touch the database, integrity-protected with HMAC-SHA256 — and can never be read back through the MCP interface. Not by the AI, not by anyone.
Same-server verification
Before a single file is pushed, AlphaBridge proves with a probe file and a cryptographic signature that the FTP/SFTP target really is the server your site runs on. A redirected or spoofed host fails the check — SFTP host keys are pinned by SHA256 fingerprint.
Allowlists, not blocklists
After an independent security review we rebuilt the guard rails: what is not explicitly allowed is refused. Every object operation checks the permission for that object, and every state-changing tool ships disabled until you switch it on.
Every layer, spelled out.
No vague promises — this is what the shipped plugin actually does.
- OAuth 2.1 with PKCE. Claude connects via a login-protected consent screen on your own site — standard discovery, no token copying, no account with us.
- Hashed tokens, scoped and revocable. Tokens are stored only as hashes, can be limited to read-only or content-only, given an expiry, rotated in one click and revoked at any time.
- Permission checks per tool and per object. Every call runs as a real WordPress user — and checks the capability for the specific post, file or term it touches, not just a global role.
- Layered secret protection. Protected meta keys, credential-pattern keys (api_key, token, secret …) and options are simply not reachable over MCP — reads and writes are refused before any tool logic runs.
- SQL guard with a positive allowlist. Database tools (Pro) accept only allowlisted statement types, block system schemas and dangerous functions, and close comment, CTE and PREPARE bypasses — verified by 78 dedicated unit tests in every build.
- Atomic deploy with automatic rollback. Full-site deploys unpack to a staging folder, verify every file, keep a backup and swap by rename. An interrupted upload never touches the live site; a failed swap rolls back on its own — verified by 14 dedicated tests.
- Guards everywhere. Path-traversal guards, SSRF-safe URL fetching with re-validated redirects, a fixed abuse rate limit and a global read-only switch.
- A full audit log. Every tool call is recorded with user, tool, time and result — allowed or denied. When a client asks what happened, you can answer.
Reviewed, rebuilt, and tested on every build.
Before launch, we put the Pro toolset through an independent security review — and it failed. The review demonstrated real bypasses in the original guard rails. We did not patch around them: we rebuilt the security model from blocklists to positive allowlists, added object-level permission checks throughout, and turned every risky tool off by default.
The bypasses from that review are now regression tests. Every release must pass 78 SQL-guard tests, 14 atomic-deploy tests and a set of packaging hygiene gates before a ZIP is even built. Security here is not a marketing section — it is the build pipeline.
Built for people who answer for other people's sites.
If you manage client sites, "the AI did something" is not an acceptable incident report. AlphaBridge keeps you in the position to answer precisely.
Per-client connections
One scoped, expiring token per client and per tool set. Rotate or revoke without touching anything else.
Evidence on demand
The audit log answers who did what and when — for every site, every call, allowed and denied alike.
Blast-radius control
Tool groups stay off until you enable them, dry-run previews come before mass changes, and deploys can only reach verified servers.
Security is the product. The rest is convenience.
Start with the free plugin and see the audit log, scopes and safe defaults for yourself — then upgrade when your clients depend on it.
Start free Compare plans