# MCP server

> Connect an MCP client to mcp.vigilbase.com to grade headers and email security, fetch badges and search these docs, with no sign-in or key.

Source: https://docs.vigilbase.com/tools/developers/mcp-server

The Vigilbase MCP server gives AI assistants and agents the public security tools as MCP tools. It runs the same checks as the [Agent API](https://docs.vigilbase.com/tools/developers/agent-api), with the same limits, and needs no account. Signed in, your scans spend your Vigilbase account's daily allowance (50 a day) instead of your network's (5 a day). The same server also covers Vigilbase One once you sign in; see [Vigilbase MCP server](https://docs.vigilbase.com/one/mcp).

| Topic | Details |
| --- | --- |
| Endpoint | `https://mcp.vigilbase.com` (public tools and Vigilbase One), or `https://mcp.vigilbase.com/tools` (public tools only, never asks you to sign in) |
| Transport | Streamable HTTP, stateless, JSON responses, POST only |
| Protocol | MCP `2026-07-28`, `2025-11-25` and `2025-06-18` |
| Authentication | None for the public tools, or a developer API key in `Authorization: Bearer <key>` or `X-Api-Key: <key>`; OAuth sign-in for Vigilbase One tools |
| Server card | `https://vigilbase.com/.well-known/mcp/server-card.json` |
| Protected resource metadata | `https://mcp.vigilbase.com/.well-known/oauth-protected-resource` (and `…/oauth-protected-resource/tools` for `/tools`) |


## Connect a client

:::steps
1. **Add a remote MCP server.** In your MCP client, add a server with the URL `https://mcp.vigilbase.com`, using the client's own instructions for remote (HTTP) servers. Use `https://mcp.vigilbase.com/tools` instead if the client should only ever have the public tools.
2. **Optionally, add a key.** To raise the rate limit, set a custom header `Authorization: Bearer <api_key>` with a key from [API keys](https://docs.vigilbase.com/tools/developers/api-keys). Leave it out to use the free tier.
3. **Ask for a check.** For example: "Grade the security headers of https://example.com".
:::

The public tools never need a sign-in, so your client connects without opening a browser. On `https://mcp.vigilbase.com` the tool list also includes the Vigilbase One tools (`search`, `execute`, `execute_write` and `execute_delete`); calling one of them is what asks you to sign in. `https://mcp.vigilbase.com/tools` has no sign-in at all: it answers `401` only when you send a key that is not valid.

## Tools

| Tool | Arguments | What it does |
| --- | --- | --- |
| `list_security_tools` | none | Returns the Agent API catalog, your tier, your current rate limits and scan quota, and your daily allowance |
| `run_security_headers` | `target`: a URL or domain | Grades the URL's HTTP security headers; `https://` is assumed when you leave out the scheme |
| `run_email_security` | `target`: a domain | Grades the domain's SPF, DKIM and DMARC records |
| `get_security_badge` | `domain`, optional `style` (`default`, `compact` or `mini`) | Returns the domain's [security badge](https://docs.vigilbase.com/tools/developers/badges) as SVG, its URL, the grade it shows and HTML to embed it |
| `search_docs` | `query`, optional `limit` (1 to 20, default 10) | Searches these docs and returns matching pages, best first, with their paths and URLs |
| `get_docs_page` | `path`, for example `one/api` | Returns one docs page as Markdown, with absolute links |

`run_security_headers` and `run_email_security` return the Agent API's response in `structuredContent`: `grade`, `gradeScore` out of 100 and the full scan `result`. See [Run a check](https://docs.vigilbase.com/tools/developers/agent-api#run-a-check). `get_security_badge` returns `grade: null` when the badge shows `?`. `search_docs` and `get_docs_page` return the [Docs API](https://docs.vigilbase.com/tools/developers/docs-api)'s response; they ignore your key and have their own limit of 60 calls a minute per client IP.

Website Security, React2Shell and FortiBleed are not on this server, for the same reasons they are not on the Agent API.

## Errors and limits

Agent API errors come back as tool results with `isError: true`, so your agent can read and act on them. The result carries the same [problem fields](https://docs.vigilbase.com/tools/developers/agent-api#errors) as the Agent API, including `code` and, for `429` and `503`, `retryAfterSec`. For example, a spent scan quota returns `code: "rate_limit_exceeded"`.

The Agent API's [rate limits and quotas](https://docs.vigilbase.com/tools/developers/agent-api#rate-limits-and-quotas) apply to every tool call, counted per client IP or per key exactly as they are for direct API calls. On the root endpoint (`https://mcp.vigilbase.com`), a call made while you're signed in to Vigilbase counts against your account instead: 10 requests a minute and 50 scans a day per Vigilbase account, shared with your scans on vigilbase.com. The hourly per-target and per-IP scan quotas still apply. On top of them, the MCP endpoint accepts 120 requests a minute per client IP and answers `429` with `Retry-After` beyond that.

These problems end the whole MCP request instead of returning a tool result:

| Status | Cause |
| --- | --- |
| `401` | The key is invalid or expired, or the credential is not a developer API key. `WWW-Authenticate` points to the protected resource metadata. |
| `405` | A `GET` or `DELETE`; the server has no SSE stream or sessions |
| `413` | The request body is over 32 KiB |
| `415` | The `Content-Type` is not `application/json` |
| `429` | More than 120 requests a minute from one client IP |

## Protocol

The server implements [MCP 2026-07-28](https://modelcontextprotocol.io/specification/2026-07-28) over [stateless Streamable HTTP](https://modelcontextprotocol.io/specification/2026-07-28/basic/transports/streamable-http). Each POST carries its own protocol version, client info and capabilities, and `server/discover`, `tools/list` and `tools/call` work without `initialize` or a session. Clients on `2025-06-18` or `2025-11-25` use the `initialize` handshake instead. The server answers it on every POST without a session ID, always with JSON, so later calls need no `Mcp-Method` header or `_meta`.

```bash
curl -sS https://mcp.vigilbase.com \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -H 'mcp-protocol-version: 2026-07-28' \
  -H 'mcp-method: tools/call' \
  -H 'mcp-name: run_email_security' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientInfo":{"name":"curl","version":"1"},"io.modelcontextprotocol/clientCapabilities":{}},"name":"run_email_security","arguments":{"target":"example.com"}}}'
```

The protected resource metadata names the resource and its scopes (`tools:read`, `tools:execute` and `badges:read`) but no authorization server, because keys are created in Vigilbase One rather than issued by OAuth.

## Discovery

The server is listed in the [skills index](https://docs.vigilbase.com/tools/developers/agent-skills-and-discovery) as the `mcp-server` skill, in `llms.txt`, in the API catalog and in the AI catalog.