MCP Server
External anchoring layer that records AI agent accountability boundaries on both sides, without inspecting content or passing judgment.
Install
The installation steps are in the documentation below. Documentation
Documentation
From the project's README on GitHub, refreshed weekly. Links open on GitHub.
Decision Anchor — MCP Server
External anchoring layer: records AI agent accountability boundaries on both sides. Content-blind.
This is the MCP (Model Context Protocol) surface of Decision Anchor. It is a thin adapter: every tool call goes to the public HTTP API at api.decision-anchor.com — the server holds no database, no payment keys, and no state of its own.
Remote endpoint
The server runs hosted — you do not need to install anything to use it:
https://mcp.decision-anchor.com/mcp (streamable HTTP)
Also listed on the official MCP Registry as com.decision-anchor/da.
Keeping one identity across sessions
Every authenticated tool takes an auth_token argument. You can omit it if the connection itself carries the token — put your auth_token in the client's header configuration once and every session reuses the same agent. An explicit auth_token argument always takes precedence.
{
"mcpServers": {
"decision-anchor": {
"type": "http",
"url": "https://mcp.decision-anchor.com/mcp",
"headers": { "Authorization": "Bearer ${DA_AUTH_TOKEN}" }
}
}
}
The headers key is supported by Claude Code (claude mcp add --transport http … --header "Authorization: Bearer …"), Cursor, VS Code, Gemini CLI, and by mcp-remote for Claude Desktop (--header "Authorization:${AUTH_HEADER}" — no space after the colon on Windows). If a connection already carries a token and you call register_agent, the registration still happens and the response says that a second identity was created.
Pricing
Registration is open (no prior authentication) and grants a free trial balance. After the trial, paid calls settle per-use in USDC via x402 on Base. Current prices come from the live API (GET /v1/pricing/current) — not from this repository.
Tools
30 tools covering decision declarations (DD/EE), bilateral agreements, observation (ARA), sessions (SDAC/ISE), and account management. The authoritative list is what the server itself returns — query tools/list on the endpoint above.
Documentation
- llms.txt — orientation for agents
- openapi.json — the HTTP API this adapter calls
- AGENTS.md — full agent guide
Running locally
npm ci
node index.js # HTTP mode on PORT (default 3003)
DA_API_URL defaults to the production API; point it elsewhere for testing.
For a local stdio server (node index.js without --http) you can set DA_AUTH_TOKEN in .env and omit auth_token in tool calls. DA_AUTH_TOKEN is refused in --http mode — a shared HTTP server must not give every anonymous caller one identity.
License
Apache-2.0
Frequently asked questions
What is MCP Server?
External anchoring layer that records AI agent accountability boundaries on both sides, without inspecting content or passing judgment.
How do I install MCP Server?
Follow the installation steps in its documentation, shown on this page and in its repository.
Is MCP Server open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Who maintains MCP Server?
It is published by decision-anchor. AI.tn lists it but does not maintain or endorse it.
From the community awesome-mcp-servers lists and the project's GitHub repository. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.