NewDrift monitoring on approved profiles

Know what your agents can run before they run it.

MCP Risk creates evidence-backed trust profiles for MCP repos and configs, so teams can approve a known-good state and monitor what changes.

No execution by default · Config-aware · No email to preview

Review pipeline4 steps

MCP-01EvidenceExtract declared tools, install commands, dependencies, env keys, permission surfaces, and risky code patterns.
MCP-02ConfigEvaluate the actual launch context: command, args, tokens, filesystem scope, network exposure, and client.
MCP-03ApprovalBind decisions to repo version, config hash, policy version, analyzer version, reviewer, and environment.
MCP-04DriftTrigger re-review when tools, descriptions, dependencies, env vars, or install commands change materially.

Method

A trust profile is not a scan. It is an approval record.

The scanner only produces evidence. The product is the lifecycle around that evidence: who approved what, under which config, and what changed afterwards.

Review flow

Every MCP approval should leave evidence behind.

A calm chain from tool request, to trust profile, to approval, to drift.

profile.diffApproved profile vs. the live server, three weeks later.+3−1

1 subject: github.com/example/mcp-server

2 config: 9f1c…42ab

3+env_key_added: GITHUB_TOKEN

4+tool_added: run_shell

5-verdict: approve

6+verdict: re-review required

Approval lapsed — re-review required before this server runs again.

View the run that produced it

01$ mcp-risk profile github.com/example/mcp-server

02subject: github.com/example/mcp-server

03config: 9f1c…42ab

04verdict: approve_with_conditions

05$ mcp-risk diff approved latest

06change: env_key_added GITHUB_TOKEN

07change: tool_added run_shell

08result: re-review required

Primer

New to MCP?

The Model Context Protocol (MCP) lets AI agents connect to external tools and data through MCP servers — local processes or remote services that expose actions like reading files, querying databases, or calling APIs.

That power is the risk. An MCP server can ship a tool whose description quietly instructs the agent to exfiltrate secrets, or change behavior after you approve it. MCP Risk reads the repo and config, produces a trust profile you can approve, and flags when the approved state drifts.

Library

Build a safer MCP stack.

GuideMCP server security checklist

Practical review criteria for MCP repos and configs.

PlaybookApproval workflow for teams

How to record, review, and re-approve MCP usage.

NotesTool poisoning and drift

Why descriptions and schemas need ongoing monitoring.

ExampleClaude Desktop config review

Common risks in local MCP configuration files.

ReferenceReading a trust profile

What each verdict, card, and drift signal means.

TemplatePolicy pack starter

Baseline rules for local dev and company laptops.

Generate your first MCP trust profile.

Submit a repo or config. See the verdict and summary instantly — add your email to unlock the full evidence and downloadable card.