Model Context Protocol server

Give Claude Code
production eyes

Connect Asenvra to Claude Code, Cursor, or any MCP client in one command. Your agent gets eight tools to scan for secrets and mock data, trace regressions, propose diffs, and redeploy — without ever leaving the terminal.

Free tier included · No local install · Tokens hashed at rest · Code never stored

terminal
claude mcp add asenvra \
  --transport http \
  https://asenvra.com/api/mcp \
  --header "Authorization: Bearer vr_mcp_your_token_here"
8
Agent tools
0
Local installs
HTTP
Transport

Connect in three steps

Asenvra runs as a remote MCP server. There is no package to install and no daemon to babysit — Claude Code talks to it over HTTPS with your token.

01

Create an MCP token

Sign in and open MCP & Agent API in your dashboard. Name a token after the machine that will use it — "Claude Code — laptop" — and copy it. Tokens are hashed at rest and shown exactly once.

Open MCP & Agent API
your token
vr_mcp_2f8a...   # copy it now — shown once
02

Add the server to Claude Code

One command wires it up. Asenvra speaks MCP over streamable HTTP, so there is nothing to install and no daemon to keep running locally.

option A — claude mcp add
claude mcp add asenvra \
  --transport http \
  https://asenvra.com/api/mcp \
  --header "Authorization: Bearer vr_mcp_your_token_here"
option B — ~/.claude.json
{
  "mcpServers": {
    "asenvra": {
      "type": "http",
      "url": "https://asenvra.com/api/mcp",
      "headers": {
        "Authorization": "Bearer vr_mcp_your_token_here"
      }
    }
  }
}
03

Ask your agent to ship safely

Run /mcp inside Claude Code to confirm the connection, then talk to it normally. The agent picks the right tool — scanning, tracing, diffing, deploying — on its own.

verify
$ claude
> /mcp

  asenvra   connected   8 tools

> Scan github.com/me/my-app with Asenvra
  and fix the secrets it finds.

The eight tools your agent gets

Tools are feature-gated by plan tier, not by quota. Your agent only ever sees the tools your plan grants — and on Enterprise, only the ones your org's PolicyShield rules allow.

ToolPlanWhat it does
scan_projectFreeRun a full ShipCheck scan from a GitHub URL or a base64 ZIP.
get_reportFreeFetch a stored scan report by its scan id.
check_dependenciesPronpm-audit dependency check — CVEs, abandoned and hallucinated packages.
trace_issueProTrace a change back to the AI session or commit that introduced it (CodeTrace).
suggest_fixEliteReturn a structured unified diff for an issue. Never writes to your files.
run_deep_analysisEliteAI second-pass security review of the files a scan flagged.
auto_fixEnterpriseApply whitelisted deterministic fixes (env, config, safe dep bumps) — optionally as a GitHub PR.
trigger_redeployEnterpriseTrigger a Vercel or Railway redeploy once the fixes land.

Built for agents that can be wrong

An agent with write access needs brakes. Every one of these is enforced server-side — nothing the client sends can widen its own permissions.

Server-side tiering

Your plan is resolved live from Stripe on every call. A client claiming Enterprise still gets Free tools.

Runaway loop guard

Repeated write calls inside a rolling 120-second window are blocked until a human confirms.

Fix whitelist

auto_fix only touches deterministic categories — env, config, safe dependency bumps. Your logic is never rewritten.

Hash conflict check

A fix is refused if the file changed since the scan, so an agent cannot clobber your newer edits.

Full audit trail

Every tool call, caller, and outcome is logged. Elite and Enterprise can read it from the dashboard.

PolicyShield

Enterprise admins cap what MCP clients may do org-wide — read-only, suggest-only, or full auto-fix.

Wire it up in one command

Create a token, paste one line into Claude Code, and your agent can scan, trace, and fix before you ever open a browser tab.

Free tier includes scan_project & get_report · No credit card