DCL Trust Oracle is a deterministic AI audit layer, natively integrated with the Model Context Protocol (MCP), that evaluates LLM and agent outputs against configurable policies before and after action. Every verdict is written to a tamper-evident, hash-chained audit log that stores only cryptographic metadata — never raw content — enabling privacy-first, post-action forensic review. The server exposes 18 tools spanning jailbreak detection, secret and PII leak scans, crypto/trading compliance (wallet guardian, MEV and market-signal screens), and post-action forensic decode, alongside fast and strict pre-action audit tiers. Both the MCP server and a companion REST API are metered and settled per-call via the x402 micropayment protocol on the Base network (USDC), letting autonomous agents pay without any account setup.
Run: claude mcp add --transport http dcl-trust-oracle https://dcl-trust-oracle--fronesislabs.run.tools
Can I add DCL Trust Oracle as a connector in ChatGPT or Claude.ai?
Yes — it’s a remote MCP server, so no local install is needed. ChatGPT: Settings → Connectors → Advanced → Developer mode → Add connector, then paste https://dcl-trust-oracle--fronesislabs.run.tools. Claude.ai: Settings → Connectors → Add custom connector, then paste the same URL.
Menjometre is an independent observatory of Catalan public spending. This MCP server gives any LLM client read-only access to the full dataset and the statistical scoring layer built on top of it.
## Coverage
- **~19.5 M** grant records from the Registre d'Ajuts i Subvencions de Catalunya
(RAISC), 2016–present
- **~1.7 M** public procurement contracts from Dades Obertes de Catalunya
- **~496 K** beneficiary entities with normalised identities
- A **public-figure graph** linking Parlament deputies, Generalitat officials, and
the boards of publicly-funded entities
- The **Menjometre score** — a statistical indicator flagging concentration and
recurrence patterns in public spending
## What the 48 tools cover
| Domain | What you can ask |
|---|---|
| `entities` | Search, profile, network, activity timeline for any funded
organisation |
| `grants` | RAISC lookups, concentration analysis, purpose trees, year-over-year
trends |
| `contractes` | Procurement lookups, sole-source detection, fragmentation patterns |
| `organs` | Granting-body rankings and profiles |
| `xarxa` | Public-figure graph, shortest path between two people, cluster
investigation |
| `pressupostos` | Generalitat budget surface |
| `meta` | Methodology, scoring explanation, upstream attribution manifest |
## Example prompts
- *"Which ten Catalan entities received the most RAISC funds in 2024?"*
- *"Show me the concentration profile of the entity with CIF A08000143."*
- *"Which grant-giving bodies award the largest share of their budget to a single
beneficiary?"*
- *"What percentage of Ajuntament de Girona's contracts are sole-source?"*
- *"Explain how the Menjometre score is computed."*
## Response envelope
Every tool response ships with four things:
1. **Attribution** — a `sources` array citing the upstream dataset, licence
(CC-BY-4.0 / Llei 37/2007), and publisher
2. **Scope predicate** — personal-data filters enforced at the database query, not
client-side
3. **Framing** — scored responses embed a `score_meta` block with a
`not_an_accusation` disclaimer and a link to the methodology page
4. **Neutral statistical language** — the server returns "top decile", "around the
median", never accusatory framing
Safe for journalism, academic research, and civic-tech tooling.
## Install via npm
```json
{
"mcpServers": {
"menjometre": {
"command": "npx",
"args": ["-y", "menjometre-mcp"]
}
}
}
Also reachable directly over Streamable HTTP at https://www.menjometre.cat/mcp.
