Decision Log

Agent developers and Organization administrators

MCP and agents

Register scoped MCP clients, discover supported tools and resources, and preserve Actor attribution for agent activity.

Task 1

Register a client
Create an Organization-scoped MCP client and retain its one-time secret securely.

Permission boundary

mcp_clients.manage.organization

Steps

  1. 1Register the client through the API or repository-local CLI.
  2. 2Store the returned client secret in the caller's credential vault.
  3. 3Send the client identifier and Bearer secret on MCP requests.
  4. 4Revoke the client when its owner or use case changes.

Registration surfaces

GraphQL mutation: registerMcpClient
pnpm decision-log mcp clients register
Endpoint: https://www.decisionlog.ai/api/mcp

Registration and revocation are audited. Plaintext client secrets are returned only at creation.

Task 2

MCP tools
Discover Decision Log as a Streamable HTTP MCP server before invoking permission-scoped tools.

Steps

  1. 1Read the public server manifest to discover the current endpoint and required headers.
  2. 2Use OAuth protected-resource metadata when the client supports the authorization flow.
  3. 3Prefer read tools first; creation and supersession tools enforce the same permissions and audit rules as GraphQL.
  4. 4Treat extraction output as reviewable DecisionCandidates unless trusted policy explicitly allows otherwise.

Discovery

GET https://www.decisionlog.ai/.well-known/mcp.json
GET https://www.decisionlog.ai/.well-known/oauth-protected-resource/api/mcp
POST https://www.decisionlog.ai/api/mcp

The manifest is the source for the current remote URL and required transport headers.

Task 3

Audit Agent actions
Keep the registered MCP client, Agent Actor, permission decision, and resulting product action attributable.

Permission boundary

audit_events.read.organization to list raw audit events

Steps

  1. 1Register clients and Agents under the owning Organization.
  2. 2Use scoped credentials rather than a shared human session.
  3. 3Review append-only audit events for meaningful reads, writes, denials, and MCP invocation context.
  4. 4Investigate the source and supersession chain before accepting an agent-produced answer as current.

Audit access

GraphQL query: auditEvents
pnpm decision-log audit list --action mcp_tool.invoked --limit 25
MCP tool: decision_log.list_audit_events
MCP app: decision_log.render_audit_explorer

GraphQL, the repository-local CLI, and the structured MCP tool use the same permission-aware audit projection. The MCP audit explorer remains the visual review surface.

Continue with another workflow

Decision modelRecord immutable Decisions, retrieve the current answer, and preserve every change through explicit supersession.API and CLIAuthenticate scoped requests and use the public API or repository-local CLI against the same canonical operations.Sources and extractionSubmit provenance-bearing source material, run reviewable extraction, and resolve DecisionCandidates explicitly.SecurityUnderstand authentication, authorization, scoped credentials, and deployment boundaries without treating identity as permission.OperationsRun the application locally, execute the supported validation matrix, and collect evidence for a production release.
Create an accountAsk a question