MCP

The Blankstate MCP server exposes the IBF sensing API as MCP tools. Any MCP-compatible host (Claude Desktop, Cursor, Continue, custom agent runtimes) can measure content and guard its own actions against your Protocols without custom HTTP code. Today the server exposes sensing (bks_sense) and a runtime guardrail (bks_validate). Interactive Dives — where your agent runs inside a Proxy World and earns a measured StageCard — are the next surface on this transport.

Install

Add the server to your MCP host configuration:

{
  "mcpServers": {
    "blankstate": {
      "command": "npx",
      "args": ["-y", "@blankstate/mcp"],
      "env": {
        "BLANKSTATE_API_TOKEN": "bks_your_token",
        "BLANKSTATE_PROTOCOLS": "proto-xxx:1.0"
      }
    }
  }
}

Env var	Required	Description
`BLANKSTATE_API_TOKEN`	yes	API token from the Atlas API Dashboard.
`BLANKSTATE_PROTOCOLS`	one of protocols/metrics	Comma-separated protocol ids (e.g. `proto-xxx:1.0,proto-yyy:2.0`).
`BLANKSTATE_METRICS`	one of protocols/metrics	Comma-separated metric ids for weighted, multi-protocol aggregation.
`BLANKSTATE_MODE`	no	Default enforcement mode for `bks_validate`: `block` · `feedback` · `audit`.
`BLANKSTATE_API_URL`	no	Defaults to `https://ibf.blankstate.ai`.

Available tools

Tool	Description
`bks_sense`	Measure any content against a configured Protocol or Metric. Returns score, per-metamarker resonance, glass-box evidence, and fidelity.
`bks_validate`	Validate a proposed agent action (`tool_name` + `tool_args`) against the configured Protocols/Metrics. Returns `ALLOWED`, `BLOCKED` (with evidence), or `FEEDBACK` per the configured mode — the runtime guardrail.
`bks_status`	API connectivity, authentication, ICS balance, and the configured protocols/metrics.

Usage in an agent

Once the MCP server is running, the agent calls the tools naturally. Sensing — the agent measures content:

Sense this advisory call transcript with the advisory-quality metric and summarise the fidelity and key metamarkers.

The agent calls bks_sense, receives the metamarker set, and composes the summary. Guardrail — the agent checks an action before it runs:

Before executing this command, validate it.

The agent calls bks_validate with the tool name and arguments; a BLOCKED result suspends the action with the matched nuances and evidence, so unsafe operations are caught before they execute.

Notes

All measurement runs server-side; no proprietary models or algorithms ship in the package.
Tokens are minted and metered (ICS) via the Atlas API Dashboard.

Overview

Packages

Install

Available tools

Usage in an agent

Notes

​Install

​Available tools

​Usage in an agent

​Notes

Install

Available tools

Usage in an agent

Notes