Trust MCP Server
Connect AI assistants to the Rasepi Trust Engine over the Model Context Protocol. The rasepi-mcp-trust server exposes the trust gate, AI app registry, policies, and compliance reports as 13 focused tools.
Overview
The Trust MCP server lets an assistant evaluate sources before it uses them, register and key AI applications, inspect policies, and pull compliance reports. Every decision is tenant-scoped and recorded in the decision ledger for audit.
The trust gate, AI app registry, key minting, policies, stats, and compliance reports.
Partition retrieval sources into allowed, warned, and blocked before the model reads them.
Each decision lands in the ledger with its evidence so you can reconstruct any answer.
Standard stdio transport works with Claude Desktop, VS Code, and any MCP client.
Install
cd developers/mcp
npm install
npm run build
For development without building, run the Trust server directly:
npm run dev:trust
Authentication
The server authenticates with a Bearer token passed via environment variables:
RASEPI_API_URL=http://localhost:5000
RASEPI_API_TOKEN=dev-token-{tenantId}:{userId}Dev tokens use the format dev-token-{'{'}tenantId{'}'}:{'{'}userId{'}'}. In production, use a JWT from the OAuth flow.
Claude Desktop Configuration
Add the Trust server to claude_desktop_config.json, pointing at bin/trust.js:
{
"mcpServers": {
"rasepi-trust": {
"command": "node",
"args": ["path/to/developers/mcp/dist/bin/trust.js"],
"env": {
"RASEPI_API_URL": "http://localhost:5000",
"RASEPI_API_TOKEN": "dev-token-tenant:user"
}
}
}
}Trust & Governance Tools (13)
trust_evaluateEvaluate sources (allow/warn/block/escalate) with reasons
trust_filter_sourcesRAG gate: partition sources into allowed / warned / blocked
trust_get_sourceDecision-free trust metadata for one source
trust_log_usageRecord which sources an AI actually relied on
trust_get_decisionFetch a past decision with its evidence (audit trail)
list_ai_appsList registered AI applications
get_ai_appGet an AI app with its keys and sources
create_ai_appRegister a new AI application
mint_ai_app_keyMint an expiring Trust API key bound to an app
list_trust_policiesList the tenant AI-use policies
trust_statsTrust dashboard aggregate over a window
list_trust_reportsList available compliance reports
get_trust_reportRun a compliance report (date + per-app filters)