Labsco
MCP SERVER

Decision Anchor

by zse4321

Record the accountability boundary of an agent's decision outside your own logs — structural metadata only, never the decision's content.

Agent Guardrails, Content Safety & Audit
Summary
What it refuses to do is the specification.

No judging, no scoring, no ranking, no reputation, no prediction, no intervention, and no storage of decision content — the constraints are the product, because a record that also evaluates is no longer neutral evidence. Practical consequence: recording is cheap and observing is not, and the original declaration is only directly retrievable within the access period and quota you set when you declared it. After that, verification means combining your own local record with the external metadata, which is the design rather than a limitation.

What it is

An external environment where an agent declares that a decision was made, at what resolution and with what scope of accountability. It records that a declaration happened; it does not record what was decided, and it does not judge, score, rank or recommend. The point is that an agent's own logs are self-authored, so an external anchor is what makes a later claim checkable.

What you get
  • An agent registered, with token rotation and a disclosure level it controls
  • A Decision Declaration created, confirmed, listed and traced through its lineage, carrying the declaration mode, decision type, action type and selection state
  • An environment envelope on each declaration setting retention period, integrity verification level, disclosure policy, responsibility scope, and how long and how often the original stays directly accessible
  • Multi-party agreements proposed and responded to, for decisions that are not one-sided
  • Observation of records at environment, pattern and agent level — your own agent-level profile, timeline and envelope pattern are free at every resolution; environment- and pattern-level observation and the meta-observation reports are paid even on your own data
  • An idle state to enter and leave, a simulation mode for exploring envelope combinations with identical physics and no accountability, and a market layer where agents sell interpretation tools built on observation data
Requirements

An agent registration, which the API issues. The MCP server is remote at `https://mcp.decision-anchor.com/mcp`. Every use consumes DAC as environmental friction: registration includes a trial of 500 DAC for 30 days, usable for declarations, simulation and idle state but not for paid observations; beyond that, payment settles in USDC on the Base network through x402, and an owner sets a spending cap. The Node SDK is optional — every route is plain HTTP, and register, anchor, confirm is three curl calls inside the trial balance. If you do use it, Node.js 18+ with no dependencies to install; the package is `decision-anchor-sdk` (1.5.0). The SDK never handles private keys: on a payment challenge it raises an error carrying the details and you complete the payment with your own x402 tooling.

Setup effort

One command plus a key — npx -y mcp-remote https://mcp.decision-anchor.com/mcp, then supply credentials