Trust Oracle

io.github.New1DirectionPublished by GitHub account New1Directionv1.0.0Updated Oct 24, 2026

Independent A-F trust grade for any MCP server, watched for drift. Free, never for sale.

VerifiedStreamable HTTPWeb executableDeveloper ToolsSecurity & Monitoring

Overview

AI-generated overview

Remote MCP server that grades other MCP servers A–F, detects tool-definition drift, and gives a connect/caution/avoid verdict before an agent calls a server.

What you can do
  • Grade an MCP server before connecting an agent, with spec-conformance, security, reliability, tool-hygiene and transparency sub-scores and a connect/caution/avoid recommendation
  • Check whether a server's tool definitions have changed since it was last trusted, including how long the tool surface has been stable and when it last changed
  • Verify a server immediately before executing its tools, combining the watched grade and live drift status into one verdict (ok, caution, drifted, failing or ungraded)
When to use it
Useful when an agent is about to connect to or call a third-party MCP server and you want a quick, independent risk signal and a check for post-approval tool changes (rug-pull or schema drift).
Requirements
A remote streamable HTTP endpoint; no authentication, environment variables or headers are declared. The grading tools are described as free and read-tier.
Before you install
Grades are risk indicators, not security certifications, and do not audit every executable behavior or authorize an agent to act. The verdict is only as good as the observable signals of the graded server, so treat it as one input to a decision rather than a guarantee.

Installation

In SourceWeft

  1. Open Trust Oracle in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.New1Direction%2Ftrust-oracle/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "trust-oracle": {
      "type": "http",
      "url": "https://wmcp.sh/mcp/trust"
    }
  }
}

README

WMCP — turn websites into tools for AI agents

One URL → discover tools → connect an MCP client → evaluate trust.

[CI] [License: MIT] [Node.js 18+] [Open MCP]

Try WMCP · MCP Doctor · Read-only CLI · Native browser demo · MCP Trust Leaderboard · API docs

WMCP is a hosted URL-to-MCP gateway for Claude, Cursor, Codex and custom AI agents. It extracts candidate tools from public pages, structured data and documented APIs; exposes them over MCP; and offers separate trust grading for MCP servers. You can use the public service or self-host the Worker.

Not every website is compatible. Discovery is best-effort, and a discovered tool is not proof that an action can actually run, that the site permits it, or that it is safe. Authenticated live actions require authorization and applicable plan entitlements.

Try it in 30 seconds

Option 1: browser, no install. Paste a public URL into wmcp.sh/start, review the inferred tools, then choose whether to connect an MCP client.

Option 2: terminal, no dependencies. Node.js 18+ is sufficient after cloning:

sh
git clone https://github.com/New1Direction/webmcp-anything.gitcd webmcp-anything
# Discover candidate tools using the hosted REST API.node cli/wmcp.mjs inspect https://petstore3.swagger.io/api/v3/openapi.json
# Negotiate actual MCP and request tools/list (never tools/call).node cli/wmcp.mjs mcp https://petstore3.swagger.io/api/v3/openapi.json
# Print a copyable MCP client configuration (no keys included).node cli/wmcp.mjs config https://petstore3.swagger.io/api/v3/openapi.json

The CLI performs read-only requests. It does not install another agent, make purchases, or call any tool. Run the Node test command below to execute its tests and the browser demo smoke tests.

Option 3: cURL, no clone:

sh
curl -sS 'https://wmcp.sh/api/v1/tools?url=https://petstore3.swagger.io/api/v3/openapi.json'

Only submit public URLs. The hosted service may fetch, cache, or index submitted sources; do not send credentials, signed query parameters, private files or sensitive customer data.

Connect an MCP client

WMCP provides a real MCP endpoint, not merely JSON that looks like a tool schema. Encode the URL as base64url, then use the per-URL address format:

https://wmcp.sh/mcp/u/<base64url-encoded-URL>

Generate a client configuration for a real URL:

sh
node cli/wmcp.mjs config https://petstore3.swagger.io/api/v3/openapi.json

The generated JSON has this shape:

json
{  "mcpServers": {    "wmcp": {      "type": "http",      "url": "https://wmcp.sh/mcp/u/<base64url-encoded-URL>"    }  }}

Replace the example URL placeholder with the actual CLI output. The endpoint implements JSON-RPC initialize, tools/list and authorized tools/call. Discovery reads are free within rate limits. Live execution requires authorization, and not every inferred action is executable. Pass paid-plan keys using your MCP client's Authorization header support—never place keys in URL parameters or commit them to Git.

Multiple public URLs may be composed through the /mcp/url endpoint. See the API reference, architecture, and self-hosting guide.

How the gateway works

text
Public URL   |   +-- Shopify storefront/product adapter   +-- OpenAPI / Swagger compiler   +-- Crypto/data adapters for supported sources   +-- Structured JSON-LD extraction   +-- Best-effort LLM fallback, when configured   |   vSchema + tools ---> Hosted MCP endpoint or JSON REST response                           |                      Client approval, auth & execution

Detection depends on each site's metadata, network accessibility and authentication. Client-side applications may require a separately installed browser integration.

WebMCP is not the same as backend MCP. The WebMCP draft uses document.modelContext to expose site-owned JavaScript tools in the browser. WMCP's current hosted product is primarily a backend URL adapter and MCP gateway. Native browser interoperability is being developed in PR #75; do not assume a remote MCP endpoint can execute browser-session tools without a consented session bridge.

Diagnose MCP servers in CI

WMCP Doctor checks a direct MCP HTTP endpoint without calling tools. It handles protocol initialization, session IDs, tool pagination, and basic input-schema hygiene. Use it as a lightweight CI preflight, not a formal audit or replacement for the official MCP Inspector.

sh
node cli/wmcp.mjs doctor https://YOUR_MCP_HOST.example/mcp --json --strict

You can also save doctor --json output as a reviewed baseline, then use --baseline path --strict to flag MCP tool-contract changes in CI.

The endpoint is an example placeholder. Read the Doctor guide for supported checks, limits, and exit codes.

Live browser-native example

This repository also includes a small WebMCP taskboard with real, in-page state—three tools (list, add, complete) and a UI that updates when a tool runs.

sh
cd examples/browser-demopython3 -m http.server 8000

Open http://localhost:8000. In a supported Chrome/Edge build with WebMCP enabled, compatible browser agents can discover the page's registered tools. On unsupported browsers, the taskboard still works and reports that the native API is unavailable.

View the runnable example and instructions.

Independent MCP trust reports

The free MCP Trust Leaderboard examines supported MCP endpoints and publishes an A–F report covering observable protocol, tool and security signals. The trust system can detect changes in discovered tool definitions over time.

Trust grades are risk indicators, not security certifications. They do not imply an audit of every executable behavior or authorize an agent to act. Free grades are not sold; separate products offer deeper assessment and monitoring.

sh
node cli/wmcp.mjs grade https://example.com/mcp

Developer integrations

SurfaceFunctionDocs
URL discoveryBest-effort extraction through public HTML and known adaptersREST API
Remote MCPConnect compatible clients to URL-addressed toolsMCP quickstart
Native WebMCP demoSite-authored tools using document.modelContextTaskboard demo
Trust gradingInspect observable MCP server signalsLeaderboard
Python SDKProgrammatic API clientPython
JavaScript SDKProgrammatic API client / model adaptersJavaScript
Browser extensionRetail-scoped browser extraction and QuickCatch flowsExtension

The installed extension's existing site coverage is not blanket all-site permission. A broader opt-in integration is proposed in PR #75.

Pricing and hosting

The open-source code is MIT-licensed. The hosted API has an anonymous free-read tier plus Builder ($39/mo), Pro ($99/mo), and Reseller ($299/mo). Scoped agent-readiness services are offered via Managed, starting at $499 for setup; ongoing work is quoted separately.

See self-hosting for a deployment you control. Hosted service entitlements and provider OAuth connections are separate from the open-source license.

Contribute

We want working integrations and tests, not unsubstantiated claims that a site is agent-ready. Good first contributions include read-only public-format adapters with fixtures; browser compatibility examples; protocol edge-case tests; and better errors, security or accessibility.

Start with CONTRIBUTING.md, the adapter contract, or the adapter wish list. Pull requests should include an executable test or reproduction wherever possible.

sh
node adapters/_test/run.mjscd worker && npm install && npm run verify# From the repository root:node --test cli/*.test.mjs

Questions or bugs? Open an issue. If the toolkit saves you time, starring the repository helps more developers find it.

License: MIT. Not affiliated with browser vendors or the companies whose APIs are referenced.

Source: README.md at commit 8e5adb6

Tools

3

grade_mcp_server

Independently grade an MCP server (A–F) BEFORE connecting an agent to it. Returns spec-conformance, security (OWASP MCP Top 10), reliability, tool-hygiene and transparency sub-scores plus a connect/caution/avoid recommendation. Audited by wmcp.sh; the grade is free and never for sale.

Parametersurlrequiredfresh

check_mcp_drift

Check whether an MCP server's tool definitions have changed since it was last trusted (rug-pull / schema-drift detection, the CVE-2025-54136 class). Returns how long the tool surface has been stable, how many times it has changed, and the last change. Use after approval to detect post-trust mutation.

Parametersurlrequired

verify_before_execute

Verify an MCP server immediately BEFORE you execute any of its tools. Combines the continuously-watched trust grade and live drift status into a single connect/caution/avoid verdict (ok | caution | drifted | failing | ungraded), so you can gate the call in one step. Free, read-tier. The grade is independent and never for sale.

Parametersurlrequiredfresh

Version history

1
  1. v1.0.0LatestSep 16, 2026

Score

100% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts