Overwing MCP — Watchlist

MCP100 Assessment

Watchlist / Tracking

Why we’re tracking it

Follow-up value is specific to Overwing typed LLM-output guardrail verdicts: Do confidence-calibrated review results remain distinct from pass and fail when rule coverage changes?

Reviewed Oct 5, 2026 · Documentation review; complete setup workflow not independently tested

Best fit

Evaluate LLM/agent outputs against safety, privacy, compliance and custom guardrail rules.

What we know

Hosted or local MCP for pass/fail/review output evaluation, with built-in safety/outbound-message rule sets and optional account/key-management/receipt tools.

Maintenance / activity evidence

overwing-mcp 0.12.0 published 2026-10-02T03:44:30.117Z

Tracking evidence observed Oct 5, 2026. These dated signals do not establish current maintenance. Reviewed enrichment confidence: high.

Verification gaps / watch-outs

Installation and comprehensive security have not been independently audited.
Guardrail evaluation plus some account/key/plan state depending on tools. Writes: Potential account/key rotation/plan operations where enabled.
Optional API/agent key.
Sensitive access: Potential sensitive text being evaluated, API/account credentials

Guardrail service has low direct target authority but may receive sensitive text and should not be the sole enforcement boundary.

Full authority, access and evidence limits are available in Review context and Access & risk.

All limitations / Review context

Do confidence-calibrated review results remain distinct from pass and fail when rule coverage changes?

The bounded review of the publisher's documented Overwing typed LLM-output guardrail verdicts and release notices identified no disclosed material condition preventing an accurately scoped public tracking entry. The resource's operational permissions remain uncertainty; this assessment does not certify installation safety.

Guardrail service has low direct target authority but may receive sensitive text and should not be the sole enforcement boundary.
Guardrail evaluation plus some account/key/plan state depending on tools.
Optional API/agent key.
Submitted text is evaluated by Overwing; anonymous text is documented as not stored for the free keyless path.
Reviewed write actions: Potential account/key rotation/plan operations where enabled
Sensitive access: Potential sensitive text being evaluated, API/account credentials
Installation, client compatibility and comprehensive security have not been independently audited. Authentication and access depend on the configured services.
Runtime behavior, dependencies and undisclosed vulnerabilities were not independently audited.
Documented tool authority and any remote data handling must be checked before installation or use.
Tool inventory, client matrix and permissions completeness were not requirements for this Radar tracking review.
Some tools are keyless but account/key/plan operations require credentials.
A verdict is advisory unless the orchestrator enforces it.
Text submitted for evaluation may leave the local process.
Live tool surface has changed across releases.
Exact tool count has drifted across public sources; do not publish a canonical count.

Original activity observations

overwing-mcp 0.12.0 published 2026-10-02T03:44:30.117Z

Technical identity

npm/overwing-mcp

Exact implementation source: https://www.npmjs.com/package/overwing-mcp ↗
Reviewed resource type: MCP Server
Repository identity: frod27/overwing-mcp
Package identity: npm:overwing-mcp
Service endpoint: https://overwing.ai/mcp

Evidence & Sources

Evidence basis: Recorded source evidence
Setup runtime verification: Not independently runtime-tested by MCP100 for this setup workflow.

Reviewed Oct 5, 2026 · Latest source observation Oct 8, 2026

Package registry metadata ↗
overwing-mcp
Recorded package identity, distribution or download evidence.
Evidence source ↗
overwing.ai/mcp
Source material retained for this review.
Includes references whose attribution is not established.
GitHub repository ↗
frod27/overwing-mcp
Repository reference; implementation attribution is not established for every document.
Includes references whose attribution is not established.

Setup · Watchlist

Install & Agents

Publisher instructions.

Source evidence observed Oct 14, 2026.

Additional setup details required — see the specific gaps below.

Install / connect
Publisher instructions
Publisher-documented package/connection route; replace placeholders and verify host-specific syntax.
Publisher setup example · Claude Code only · Publisher instructions
claude mcp add --transport http overwing https://overwing.ai/mcp
Configure / authenticate
MCP100 setup guide
Hosted keyless discovery is possible. Local alternative `claude mcp add overwing -e OVERWING_API_KEY=... -- npx -y overwing-mcp`; preflight_check requires key and may incur cost.
Use correct transport and published host config. Never paste secrets in chat; server registration is separate from software installation.
Environment settings: OVERWING_API_KEY (optional/paid/account tools), OVERWING_AGENT_KEY (where applicable), OVERWING_BASE_URL (optional)
Check connection
Suggested check
Use beacon_sample or another explicitly keyless/read-only tool; avoid signed transactions or paid scans.
Remove / uninstall / revoke
MCP100 setup guide
If registered as `overwing-mcp` in Claude Code, remove its exact client entry with `claude mcp remove overwing-mcp`; for Cursor/Claude Desktop/Codex, remove the matching MCP entry in that host settings. This client removal is MCP100-composed, not a vendor uninstall instruction. Npx does not imply a global install; do not delete user data or revoke credentials implicitly.
Publisher install example registration name(s): overwing. The supplied removal example applies only to an existing Claude Code registration named overwing-mcp; it does not remove the differently named install example. Confirm the actual host entry before use.
Conditional host / package cleanup · Claude Code only · MCP100 composed guide · Review before use
```
claude mcp remove overwing-mcp
```

Official Documentation & Sources

What it does

Hosted or local MCP for pass/fail/review output evaluation, with built-in safety/outbound-message rule sets and optional account/key-management/receipt tools.

Capability: Security & Testing
Resource type: MCP Server
Best fit: Evaluate LLM/agent outputs against safety, privacy, compliance and custom guardrail rules.

Representative uses

Evaluate model output before sending
Check PII/confidential leakage
Apply custom plain-language rules
Inspect/verify provider receipts/guardrail metadata

Limitations

Some tools are keyless but account/key/plan operations require credentials.
A verdict is advisory unless the orchestrator enforces it.
Text submitted for evaluation may leave the local process.
Live tool surface has changed across releases.
Exact tool count has drifted across public sources; do not publish a canonical count.

Reviewed access

Access & risk

Review the exact access boundary before connecting this MCP to accounts, credentials, or sensitive data.

Authentication: Optional API/agent key.
Permissions: Guardrail evaluation plus some account/key/plan state depending on tools.
Data handling: Submitted text is evaluated by Overwing; anonymous text is documented as not stored for the free keyless path.
Cost: freemium
Risk and cautions: Guardrail service has low direct target authority but may receive sensitive text and should not be the sole enforcement boundary.

Access scope and cautions

Write actions: Potential account/key rotation/plan operations where enabled
Sensitive access: Potential sensitive text being evaluated API/account credentials
Risk level: moderate

FAQ

Based on reviewed evidence as of 2026-10-08. Watchlist tracking is not a recommendation or a security certification.

Which tasks are documented for this implementation?

The reviewed tasks include: Evaluate model output before sending; Check PII/confidential leakage; Apply custom plain-language rules; Inspect/verify provider receipts/guardrail metadata.

What setup and authentication are documented?

Recorded installation/entrypoint step: npx -y overwing-mcp. Recorded endpoint: https://overwing.ai/mcp. Recorded transport: stdio, streamable_http. Recorded environment-variable notes: OVERWING_API_KEY (optional/paid/account tools); OVERWING_AGENT_KEY (where applicable); OVERWING_BASE_URL (optional). No configuration values are supplied here. Prerequisites: MCP client. These are the documented setup details, not a claim that installation or client compatibility was independently tested. Authentication: Several evaluation/lookup tools are keyless; broader account/paid use employs Overwing key. A demo or keyless description is not independently verified read-only authority. Check the documented tool permissions rather than inferring authority from that label.

Can it modify data or perform write operations?

Documented write actions include: Potential account/key rotation/plan operations where enabled. Recorded permission boundary: Guardrail evaluation plus some account/key/plan state depending on tools.

What data-handling boundary should I consider?

Submitted text is evaluated by Overwing; anonymous text is documented as not stored for the free keyless path. Recorded sensitive-access context: Potential sensitive text being evaluated; API/account credentials.

What limitations and risks should I check before use?

Guardrail service has low direct target authority but may receive sensitive text and should not be the sole enforcement boundary. Some tools are keyless but account/key/plan operations require credentials. A verdict is advisory unless the orchestrator enforces it. Text submitted for evaluation may leave the local process. Live tool surface has changed across releases. Exact tool count has drifted across public sources; do not publish a canonical count. Runtime behavior, dependencies and undisclosed vulnerabilities were not independently audited. Documented tool authority and any remote data handling must be checked before installation or use.