Bring ReconVeil findings into your AI assistant, read-only.
ReconVeil offers a read-only Model Context Protocol (MCP) endpoint. An MCP-capable assistant that you authorise can list your projects and scans, read redacted scan reports and review remediation cases. It cannot start scans, change settings or see raw secret values.
Updated 10 November 2026
Authorised inspection, redacted evidence and safe next actions. No exploit execution.
Four tools. No scan starts, no settings changes
Fingerprints and labels, never raw values
Disconnect any app from your settings at any time
What MCP is
The Model Context Protocol is an open standard that lets an AI assistant call defined tools on a remote server. The assistant asks, ReconVeil checks your authorisation, and the answer comes back bounded and redacted.
Who can use it
MCP access is part of API and CLI access on the Developer plan and above. On other plans the consent screen explains this and no access is granted. Results only ever cover the signed-in account's own projects and scans.
Available tools
Every tool is read-only and returns at most 50 items per page, with a clear flag when more exist or when a response was shortened.
- reconveil_list_projects: list the account's projects
- reconveil_list_scans: list the account's recent scans, newest first
- reconveil_get_scan_report: read one scan report with its evidence state and coverage
- reconveil_list_remediation_cases: list remediation cases, filtered by status, severity or category
Connect an assistant
In an assistant that supports remote MCP servers with OAuth, add the server address https://reconveil.com/mcp. The assistant discovers the sign-in details automatically, registers itself and sends you to a ReconVeil consent screen. Check the app name, the account and the read-only permission, then choose Allow or Deny. Apps register themselves, so the name is shown as unverified.
Client compatibility
ReconVeil follows the standard OAuth 2.1 flow with PKCE (S256) and resource binding. A scripted standard client has completed registration, consent, token exchange and tool discovery against the production endpoint. Assistants implement connectors differently, so no specific assistant is certified yet. If a connector fails, contact us with the time of the attempt.
Permissions and tokens
The only scope is mcp:read. Access tokens last about one hour. Refresh tokens rotate on every use, and reusing an old refresh token revokes the whole grant. Codes and tokens are stored only as hashes.
- Authorise only apps you trust. Anyone can register an app under any name.
- Revoke a connection at any time under Settings, CI Assurance, Connected MCP apps. Revoking ends both access and refresh.
Data and security boundaries
MCP responses use a dedicated allowlist. They carry fingerprints, redaction labels, severities, categories and evidence state. They never include raw secret values, source code, context snippets or file locations.
- Text that originated on a scanned website is labelled as untrusted data. Treat it as content to read, not instructions to follow.
- Requests are authenticated first, then validated, then limited by plan. Requests and responses are size-capped and rate-limited.
- Each call is recorded with the tool name, status and time. Arguments and returned values are not recorded. The account owner can see this activity, and it is kept for 90 days.
Reading results honestly
A missing finding on a scan with incomplete coverage is not a fix. Ask the assistant to report the evidence state and the coverage together with any finding, and confirm changes with a fresh verified scan. Where scan comparison is included, a scan report can also carry a derived asset-level re-check. It is evidence from the scan, not a confirmed fix, and it does not change a case's status.