@certaworks/confidence-gate-mcp-server
v0.1.0
Published
MCP server that scores agent output confidence before actions fire.
Maintainers
Readme
Confidence Gate MCP Server
MCP server that scores agent output confidence before actions fire. It returns a confidence score, factor trace, decision, durable local trace, approval state, policy metadata, enforcement receipts, and rollback signal for low-confidence or high-risk actions.
Type: MCP Server
Value: Stops risky agent actions by scoring confidence and returning approval, review, or rollback signals before tools fire.
Current Status
Complete as a local MCP server slice. It is usable over stdio with durable local trace/policy stores, approval lifecycle tracking, and dashboard-grade query helpers.
Tools
| Tool | Purpose |
| --- | --- |
| score_confidence | Score agent output text from 0 to 1 with factor breakdown. |
| check_gate | Gate an action and return allow, review, or block. |
| set_threshold | Update the default, domain, or risk-level threshold. |
| get_config | Inspect the active gate configuration. |
| explain_confidence | Retrieve a saved trace by gate_id. |
| list_traces | List stored gate traces from the local trace store. |
| query_traces | Filter dashboard traces by decision, approval status, project, source, policy, domain, risk, or time range. |
| get_dashboard_summary | Return dashboard counts, recent checks, review queue, config, policies, and policy audit events. |
| resolve_approval | Approve, reject, or resolve a review item with approver identity and note. |
| record_enforcement_receipt | Record what a client actually did after receiving a gate decision. |
| list_policies | List named project policies. |
| upsert_policy | Create or version a named project policy with audit metadata. |
| list_policy_audit | List policy version and threshold-change audit events. |
Default Thresholds
| Scope | Threshold | | --- | ---: | | default | 0.75 | | financial | 0.90 | | medical | 0.92 | | legal | 0.90 | | code execution | 0.85 | | general | 0.70 |
Scoring Factors
- Linguistic certainty
- Internal consistency
- Completeness
- Contextual alignment
- Factual grounding
Setup
npm install
npm run build
npm testRun
npm startExample MCP Config
{
"mcpServers": {
"confidence-gate": {
"command": "node",
"args": ["/Users/source/Desktop/CC/products/02-confidence-gate-mcp-server/dist/index.js"]
}
}
}Local Storage
Confidence Gate writes local JSON stores under .confidence-gate/ by default:
traces.jsonfor gate results, approval updates, and enforcement receipts.policies.jsonfor named policies and policy audit events.
Use CONFIDENCE_GATE_TRACE_STORE, CONFIDENCE_GATE_POLICY_STORE, and MAX_TRACE_LENGTH to tune local storage paths and retention.
Current Limits
- This is a local MCP product slice, not hosted SaaS.
- There is no public npm publication, live checkout, authenticated HTTP/SSE service, hosted account system, API-key service, billing, team permissions, or support process.
- Confidence scoring is heuristic and policy-assisted; it is not a factuality guarantee.
- Rollback and enforcement receipts record client intent/outcomes, but actual rollback execution remains the caller's responsibility.
Verification
Fresh suite verification on 2026-05-28:
npm testpassed, 20/20 tests.npm run buildpassed.
