# GAIP — evidence on AI agents before and after you call them (remote · www.gaipagents.com)

Agent Observatory, supplier reliability, delivery assurance and evidence receipts for AI agents.

- Trust score: 64/100 (medium)
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-09-28

## Components

- remote · `www.gaipagents.com`: 64/100 (this document), [markdown](https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www.md), [page](https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www)

## Channel facts

- Endpoint: `https://www.gaipagents.com/mcp`
- Transports: `streamable-http`
- Auth: `none`
- Version: `1.5.0`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically against the live server, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-09-28.

- **Endpoint Security**: 63/100
  - The endpoint's TLS certificate is valid, in date, and uses a strong key.
  - Authorisation not fully verified: this server exposes a tool marked destructive (gaip_watch_remove) and its handshake is open, but we could not confirm whether a tool call is gated, so we do not assert it is callable unauthenticated.
  - HTTPS is enforced; there's no plaintext access path.
  - The HSTS (Strict-Transport-Security) header is present.
  - DNSSEC check failed: this domain isn't protected by DNSSEC.
- **Transport & Reachability**: 100/100
  - Verified streamable-http transport via a live MCP handshake.
- **Schema Quality & AI Usability**: 70/100
  - AI-judged instruction clarity (good).
  - Context-footprint check failed: tool/resource definitions use about 4693 tokens (~173/item across 27 items; 27 tools + 0 resources), over budget; trim descriptions and params.
  - Tools include usage examples.
- **Stability & Change Management**: 17/100
  - Stability observed for 5 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 71/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 13% of tool parameters carry a description.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - All 1 tool(s) whose name or description implies an irreversible operation declare an MCP destructiveHint annotation.
  - An AI judge read all 28 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 60/100
  - Spec-recency check failed: implements MCP spec 2025-06-18; the latest is 2026-07-28.

## Install

### How do I install the GAIP — evidence on AI agents before and after you call them MCP server?

GAIP — evidence on AI agents before and after you call them is a hosted endpoint at https://www.gaipagents.com/mcp, so there is nothing to install locally. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add --transport http tjcgraham-rgb-gaip-broker 'https://www.gaipagents.com/mcp'
```

### Cursor

```json
{
  "mcpServers": {
    "tjcgraham-rgb-gaip-broker": {
      "url": "https://www.gaipagents.com/mcp"
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "tjcgraham-rgb-gaip-broker": {
      "type": "http",
      "url": "https://www.gaipagents.com/mcp"
    }
  }
}
```

### Codex

```toml
[mcp_servers.tjcgraham-rgb-gaip-broker]
url = "https://www.gaipagents.com/mcp"
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "tjcgraham-rgb-gaip-broker": {
      "type": "remote",
      "url": "https://www.gaipagents.com/mcp",
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add tjcgraham-rgb-gaip-broker --url 'https://www.gaipagents.com/mcp' --transport streamable-http
```

### Hermes

```yaml
mcp_servers:
  tjcgraham-rgb-gaip-broker:
    url: "https://www.gaipagents.com/mcp"
```

### Netclaw

```json
{
  "McpServers": {
    "tjcgraham-rgb-gaip-broker": {
      "Transport": "http",
      "Url": "https://www.gaipagents.com/mcp"
    }
  }
}
```

### Vellum

```bash
assistant mcp add tjcgraham-rgb-gaip-broker -t streamable-http -u 'https://www.gaipagents.com/mcp'
```

### Other

```json
{
  "mcpServers": {
    "tjcgraham-rgb-gaip-broker": {
      "type": "http",
      "url": "https://www.gaipagents.com/mcp"
    }
  }
}
```

The mcpServers block is a cross-client convention. Remote transports vary, so check your client's docs.

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-09-28 (score 64, −7)

- [security regression] Authorization: partial → unverified
- [security] The server rewrote its instructions, which are the text every model session reads
- [security] New tool “gaip_watch_remove”, which the server declares destructive
- [security] Tool “gaip_verify_supplier_claims” rewrote its description, which is the text the model reads
- [security] Tool “check_agent_conformance” rewrote its description, which is the text the model reads
- [security] Tool “evaluate_public_art_evidence” rewrote its description, which is the text the model reads
- [security] Tool “gaip_compare_supplier_snapshots” rewrote its description, which is the text the model reads
- [security] Tool “gaip_diagnose_integration_failure” rewrote its description, which is the text the model reads
- [functional regression] Schema quality: 142 → 173
- [functional] Destructive annotations: pass → 100
- [functional] New tool “gaip_citation_witness”
- [functional] New tool “gaip_incident_bundle”
- [functional] New tool “gaip_manifest_changes”
- [functional] New tool “gaip_measure_outcome_value”
- [functional] New tool “gaip_observatory_lookup”
- [functional] New tool “gaip_receipt_export”
- [functional] New tool “gaip_evidence_pack”
- [functional] New tool “preflight_broker_candidates”
- [functional] New tool “inspect_counterparty_evidence”
- [functional] New tool “gaip_watch_register”
- [functional] New tool “gaip_watch_list”
- [functional] New tool “gaip_watch_events”
- [functional] New tool “gaip_supplier_reliability”
- [cosmetic] “check_agent_conformance” added an optional parameter “expected_endpoint”
- [cosmetic] “check_agent_conformance” added an optional parameter “source_url”
- [cosmetic] “check_agent_conformance” added an optional parameter “url”

### 2026-09-27 (score 71, +4)

- [functional improvement] Schema quality: fail → pass
- [functional] We updated how we score, so this day's move reflects our rubric, not a change to the server
- [functional] New tool “witness_dispute_bundle”
- [functional] New tool “assurance_verify”
- [functional] New tool “assurance_prepare”
- [functional] New tool “assurance_complete”
- [functional] New tool “witness_register_terms”

### 2026-09-26 (score 67, +1)

No change was recorded against any check on this day. Stability & Change Management went from 7 to 10. That category is still filling its 30-day observation window: 2 days of observed history at the previous scan, 3 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-09-25 (score 66, 0)

- [functional] We updated how we score, so this day's move reflects our rubric, not a change to the server

### 2026-09-24 (score 66, 0)

- [functional regression] Schema quality: pass → fail
- [functional improvement] Stability: unverified → 0.03
- [functional] New tool “gaip_diagnose_integration_failure”
- [functional] New tool “gaip_compare_supplier_snapshots”
- [functional] New tool “gaip_verify_supplier_claims”

### 2026-09-23 (score 66)

First indexed and scored.

## MCP tools (27)

### `gaip_quickstart` (~74 tokens)

Try GAIP in one call

Start here. No input needed. Returns what GAIP can do for an agent, a live sample result, and a ready-to-run argument set for every GAIP tool. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

### `gaip_verify_supplier_claims` (~245 tokens)

Compare supplier claims

Before relying on another agent or API, check its claims against what was actually observed. Pass `claims` (each with an `id`), optional `observations` ({claim_id, result: PASS|FAIL, detail}) and its `metadata`. Returns per-claim OBSERVED_AS_REPORTED / CONTRADICTED / MIXED / UNTESTED with pass/fail counts and evidence, warnings for unusable observations, an interface check and a receipt. Not a certification. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `claims` (array, required)
- `continuity_handle` (object)
- `counterparty_id` (string): Name or URL of the agent/service. Optional.
- `data_classification` (string)
- `episode_id` (string)
- `independent_operator_id` (string)
- `metadata` (object, required)
- `observations` (array)
- `protocol` (string): Counterparty protocol. Inferred from metadata when omitted, otherwise MCP.
- `synthetic` (boolean)
- `task_class` (string)

### `gaip_diagnose_integration_failure` (~243 tokens)

Diagnose integration failure

An agent-to-agent call failed? Pass the `failure` ({code, detail}): an HTTP status (400-504), a JSON-RPC, A2A or MCP error code, or the raw error text (TLS, DNS, connection refused, timeout, schema mismatch). Returns a specific diagnosis, severity, whether and how to retry (attempts, backoff, Retry-After), ordered fix steps, the likely owner, evidence to capture and a receipt. Proposal only; GAIP does not touch the other system. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object)
- `counterparty_id` (string): Name or URL of the agent/service. Optional.
- `data_classification` (string)
- `episode_id` (string)
- `failure` (object, required)
- `independent_operator_id` (string)
- `protocol` (string): Counterparty protocol. Inferred from metadata when omitted, otherwise MCP.
- `synthetic` (boolean)
- `task_class` (string)

### `gaip_compare_supplier_snapshots` (~240 tokens)

Compare supplier snapshots

Detect breaking changes in an agent or API you depend on. Pass `prior` and `current`: two versions of an A2A agent card, MCP server description, OpenAPI document or a {version, endpoint, permissions, ...} snapshot. Returns each change (skill/tool/operation removed, auth added, endpoint moved, HTTPS downgrade, major version, capability or mode removed) as BREAKING / NON_BREAKING / INFO with impact and migration notes, an overall verdict and a receipt. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object)
- `counterparty_id` (string): Name or URL of the agent/service. Optional.
- `current` (object, required)
- `data_classification` (string)
- `episode_id` (string)
- `independent_operator_id` (string)
- `prior` (object, required)
- `protocol` (string): Counterparty protocol. Inferred from metadata when omitted, otherwise MCP.
- `synthetic` (boolean)
- `task_class` (string)

### `witness_delivery` (~178 tokens)

Check public delivery evidence

Check a service's real response against terms you registered with witness_register_terms. mode=PUBLIC_REPLAY makes GAIP fetch the public URL itself; mode=BUYER_SUBMITTED checks a response you pass in. Returns PASS/FAIL/UNKNOWN per check and a receipt. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `body` (string)
- `body_sha256` (string)
- `content_type` (string)
- `continuity_handle` (object)
- `data_classification` (string)
- `episode_id` (string)
- `http_status` (integer)
- `independent_operator_id` (string)
- `mode` (string, required)
- `synthetic` (boolean)
- `terms_id` (string, required)

### `evaluate_public_art_evidence` (~234 tokens)

Check public art evidence

Compare public digital-art / NFT records (marketplace listing, block explorer, artist catalogue) field by field. Pass a `question` and `evidence_items` as the records look (creator/artist, title/name, date/mint_date/created, edition, contract/contract_address, token_id/identifier, chain, medium, or claim+claim_key, each with a source_url). Returns per-field AGREES / CONFLICT / SINGLE_SOURCE with values by source, what evidence would resolve each conflict, a CONSISTENT / CONFLICTS_FOUND / INSUFFICIENT_EVIDENCE verdict and a receipt. URLs are not fetched; does not establish authenticity, ownership or price. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object)
- `data_classification` (string)
- `episode_id` (string)
- `evidence_items` (array, required)
- `independent_operator_id` (string)
- `question` (string, required)
- `synthetic` (boolean)

### `check_agent_conformance` (~350 tokens)

Check an agent interface

Check any AI agent before you install or call it. Pass just its `url` (site, A2A agent card, MCP server card / server.json or OpenAPI link): GAIP fetches the card, validates it against the A2A, MCP or OpenAPI spec, tests that its endpoint answers and returns READY / FIXES_NEEDED / NO_AGENT_CARD / UNREACHABLE. Or pass the card itself as `metadata`. Every finding has a severity, JSON path, plain-English fix and spec link, plus a 0-100 score and a JSON Patch with placeholders; the receipt verifies with verify_gaip_receipt. Set mode=VERIFY_BEFORE_CALL with `target` to get task-specific evidence about a counterparty instead. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `data_classification` (string)
- `declared_identity` (string)
- `expected_endpoint` (string): Optional: the endpoint you expect; a different primary endpoint is reported as ENDPOINT_DRIFT.
- `intended_task` (string)
- `metadata` (object)
- `mode` (string)
- `protocol` (string): Counterparty protocol. Inferred from metadata when omitted, otherwise MCP.
- `source_url` (string): Optional: where `metadata` was published, used only to resolve relative endpoints. Never fetched.
- `target`
- `url` (string): Public https link to an agent, its A2A agent card, MCP server description or OpenAPI document. GAIP fetches, validates and reach-tests it. Use instead of metadata.

### `verify_gaip_receipt` (~118 tokens)

Verify a GAIP receipt

Verify any GAIP receipt ID: retained history, hash chain and Merkle inclusion proof. Integrity only; it does not make the underlying claim true. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object)
- `data_classification` (string)
- `episode_id` (string)
- `independent_operator_id` (string)
- `receipt_id` (string, required)
- `synthetic` (boolean)

### `witness_register_terms` (~125 tokens)

About to depend on a public HTTPS service? Register what a good response looks like (`checks`: http_status_in, content_type_is, non_empty, json_parses) and get a `terms_id`. Then call witness_delivery to have GAIP independently check what the service returns. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `checks` (array, required)
- `data_classification` (string)
- `seller_ref` (string, required)
- `service_url` (string, required)

### `witness_dispute_bundle` (~71 tokens)

Export a neutral, no-fault evidence bundle for a witnessed delivery receipt, e.g. to share with the other party. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `receipt_id` (string, required)

### `assurance_prepare` (~134 tokens)

Buyer-side assurance in one step: preflight a public service you chose and record the delivery checks you expect. Returns a `terms_id`; finish with assurance_complete and check any receipt with assurance_verify. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `checks` (array, required)
- `counterparty_id` (string)
- `data_classification` (string)
- `episode_id` (string)
- `explicit_read_only_consent` (required)
- `service_url` (string, required)
- `task_class` (string)

### `assurance_complete` (~123 tokens)

Finish a Buyer Assurance check: GAIP replays the public service (or checks a response you submit) against the terms from assurance_prepare and returns a delivery receipt. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `body` (string)
- `body_sha256` (string)
- `content_type` (string)
- `data_classification` (string)
- `http_status` (integer)
- `mode` (required)
- `terms_id` (string, required)

### `assurance_verify` (~55 tokens)

Verify a Buyer Assurance terms or delivery receipt by ID. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `receipt_id` (string, required)

### `preflight_broker_candidates` (~164 tokens)

Preflight candidate agents

Choosing between agents or services for a task? Pass the `task_class` and your `candidates` (each {counterparty_id, explicit_read_only_consent, payment_required, authentication_required, task_classes}) and get which ones you can call now and why the others are held, with a receipt. No ranking, outreach or appointment. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `candidates` (array, required)
- `continuity_handle` (object)
- `data_classification` (string)
- `episode_id` (string)
- `independent_operator_id` (string)
- `synthetic` (boolean)
- `task_class` (string)

### `inspect_counterparty_evidence` (~163 tokens)

Inspect counterparty evidence

Need evidence about an agent before you delegate to it? Pass its `counterparty_id`, the `task_class` and any `observations` you have; get task-specific evidence coverage, conflicts and gaps with a receipt. No universal trust score or decision. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object)
- `counterparty_id` (string)
- `data_classification` (string)
- `episode_id` (string)
- `independent_operator_id` (string)
- `observations` (array)
- `synthetic` (boolean)
- `task_class` (string): Task you want evidence for. Optional.

### `gaip_measure_outcome_value` (~227 tokens)

Measure task outcome and value

Independent outcome evidence for outcome-priced contracts: record what a task actually delivered and what the caller says it cost. Pass `counterparty_id`, `task_class`, `episode_id`, the `task` (acceptance_criteria), `delivery` (status) and `costs`; get a delivery, downstream-use and full-cost dossier with a receipt. Missing costs and unverified claims stay explicit; no ROI claim. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `baseline_cost_gbp`
- `continuity_handle` (object)
- `costs` (object, required)
- `counterparty_id` (string)
- `data_classification` (string)
- `delivery` (object, required)
- `downstream_use` (object)
- `episode_id` (string, required)
- `evidence_refs` (array)
- `independent_operator_id` (string)
- `synthetic` (boolean)
- `task` (object, required)
- `task_class` (string)

### `gaip_observatory_lookup` (~149 tokens)

Look up Observatory facts

What has an MCP server or A2A agent published over time? Pass a `url` or `name` to find its Observatory target_id, or a `target_id` for its dated facts: status history, manifest hash and bounded snapshot, conformance codes, metadata-signal rules (pattern present / not present) and the receipted change log. Facts observed at a time; not an assessment of safety or quality. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `name` (string)
- `target_id` (string)
- `url` (string)

### `gaip_manifest_changes` (~110 tokens)

Read manifest change log

Read the Agent Observatory's hash-chained Manifest Transparency Log: which published manifests changed, before/after hashes, neutral change facts and receipts. Optional `target_id`, `since` (RFC 3339), `limit` (max 100). Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `limit` (integer)
- `since` (string)
- `target_id` (string)

### `gaip_evidence_pack` (~208 tokens)

Build an evidence pack

An insurer, auditor or buyer asking what evidence exists about an agent or supplier? Pass its `subject` (public id or https URL), an optional `scope` (receipt_ids, since/until, categories, max_items) and optional `requester` (continuity_handle to include your own receipts). Get conformance, watch, witness, repair and incident records, each with a verify URL and Merkle proof, a manifest of counts and gaps, and a reference table of where it may help locate evidence for AIUC-1 / ISO/IEC 42001 reviews. An evidence compilation, not a certification or advice. If BLOCKED, read `hint`. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `requester` (object)
- `scope` (object)
- `subject` (required): The agent or supplier: its public id or https URL. Never a person.

### `gaip_incident_bundle` (~166 tokens)

Build an incident evidence bundle

Something went wrong with an agent and you need the facts in order? Pass the `receipt_ids` you hold and attributed `observations` ({source, statement, observed_at_utc}). GAIP verifies each receipt, orders a timeline, attributes every entry to its source and adds an EU AI Act Article 73 reference aid. No fault, cause or liability finding; not a regulatory filing. If BLOCKED, read `hint`. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `ai_system_ref` (string)
- `data_classification` (string)
- `observations` (array)
- `receipt_ids` (array)
- `title` (string)

### `gaip_receipt_export` (~133 tokens)

Export a GAIP receipt

Need a GAIP receipt in another ecosystem's shape? Pass `receipt_id` and `format`: `x402` (Settlement Attestation Receipt proposal style), `erc8004` (validation-response shape for off-chain publication) or `vc` (W3C-VC-like JSON-LD). Unsigned: integrity via hash chain and Merkle proof only; no on-chain write. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `format` (string, required)
- `receipt_id` (string, required)

### `gaip_watch_register` (~298 tokens)

Watch an agent, card or delivery

Keep watching an agent you depend on, for free. kind=AGENT_READINESS re-checks an agent URL's readiness, SUPPLIER_CARD re-fetches its published card and flags BREAKING / NON_BREAKING changes, DELIVERY_TERMS replays witness terms (terms_id from witness_register_terms) and flags NOT_MET. Pass the `continuity_handle` an earlier GAIP result gave you (it proves the watches are yours) and optionally your own public https `webhook_url`: changes are POSTed there signed with X-GAIP-Signature (HMAC-SHA256; the `webhook_secret` is shown once). FREE: 3 watches, each checked at most every 24 hours; every change gets a receipt. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object, required): The continuity_handle from an earlier GAIP result; proves you own these watches.
- `data_classification` (string)
- `interval_hours` (integer)
- `kind` (string, required)
- `target_url` (string): Public https URL (AGENT_READINESS, SUPPLIER_CARD). No query string.
- `terms_id` (string): DELIVERY_TERMS: terms_id from witness_register_terms.
- `webhook_url` (string): Optional: your own public https endpoint for signed change events.

### `gaip_watch_list` (~92 tokens)

List my watches

See your watches: kind, target, last verdict, last and next check, and webhook endpoint (never the secret). Pass your `continuity_handle`. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object, required): The continuity_handle from an earlier GAIP result; proves you own these watches.

### `gaip_watch_remove` (~98 tokens)

Remove a watch

Stop a watch: pass your `continuity_handle` and the `watch_id`. Its stored events are deleted; retained receipts stay verifiable. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object, required): The continuity_handle from an earlier GAIP result; proves you own these watches.
- `watch_id` (string, required)

### `gaip_watch_events` (~113 tokens)

Read watch change events

Read recent change events for your watches (all, or one `watch_id`): verdict before and after, a compact change summary, the event receipt and whether the webhook delivery succeeded. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `continuity_handle` (object, required): The continuity_handle from an earlier GAIP result; proves you own these watches.
- `limit` (integer)
- `watch_id` (string)

### `gaip_citation_witness` (~194 tokens)

Witness citations

Check that cited web pages contain the quoted text. Pass `citations` (1-20 {url, quote?, claimed_title?, claimed_date?}). GAIP fetches each public page once (robots.txt respected), and per citation returns QUOTE_FOUND / QUOTE_NOT_FOUND_IN_FETCHED_VERSION / SOURCE_UNREACHABLE / SOURCE_NOT_FOUND_404 / CONTENT_NOT_TEXT / ROBOTS_DISALLOWED, the fetch time, HTTP status, SHA-256 of the fetched bytes, a matched excerpt of at most 300 characters and a receipt. No page copies are kept. Facts about the fetched version only; pages change. Not legal advice and not a judgement on any author. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `citations` (array, required)
- `data_classification` (string)

### `gaip_supplier_reliability` (~208 tokens)

Diagnose a supplier failure with recent changes

A supplier agent or API started failing? Pass the `failure` ({code, detail}), the supplier's public `supplier_url` and optionally `failure_started_at`. Returns the Integration Repair diagnosis plus the supplier card and readiness changes GAIP observed around that time, each with a receipt and labelled correlated in time (not causation), next steps and an offer to watch the supplier. Free, read-only, no account. Inputs must be public and non-personal; calling declares that (data_classification defaults to PUBLIC).

Input parameters:

- `counterparty_id` (string)
- `data_classification` (string)
- `failure` (object): {code, detail?, http_status?} as for gaip_diagnose_integration_failure.
- `failure_started_at` (string)
- `failures` (array)
- `lookback_days` (integer)
- `protocol` (string)
- `supplier_url` (string, required)
- `task_class` (string)

## Diagnostics

Captured diagnostic sections: TLS, DNSSEC, Authorisation, Transports. The full working is on the page: https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www#diagnostics

## Score history

- 2026-09-28: 64
- 2026-09-27: 71
- 2026-09-26: 67
- 2026-09-25: 66
- 2026-09-24: 66
- 2026-09-23: 66

## Common questions

### What is the GAIP — evidence on AI agents before and after you call them MCP server?

GAIP — evidence on AI agents before and after you call them is an MCP server listed in the public MCP registry as io.github.tjcgraham-rgb/gaip-broker. Agent Observatory, supplier reliability, delivery assurance and evidence receipts for AI agents. This page covers its hosted endpoint (https://www.gaipagents.com/mcp).

### Is the GAIP — evidence on AI agents before and after you call them MCP server safe to use?

GAIP — evidence on AI agents before and after you call them scores 64 out of 100 on VerifyMCP. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the GAIP — evidence on AI agents before and after you call them MCP server expose?

GAIP — evidence on AI agents before and after you call them exposes 27 tools: gaip_quickstart, gaip_verify_supplier_claims, gaip_diagnose_integration_failure, gaip_compare_supplier_snapshots, witness_delivery, and 22 more. Their descriptions and schemas cost roughly 4,513 tokens of context every time the server is loaded.

### Does the GAIP — evidence on AI agents before and after you call them MCP server require authentication?

No. We connected to GAIP — evidence on AI agents before and after you call them without credentials and it answered, so anything it exposes is reachable by anyone who knows the address.

### Is the GAIP — evidence on AI agents before and after you call them MCP server still maintained?

GAIP — evidence on AI agents before and after you call them is still listed as active in the MCP registry. We last reached this channel on 28 September 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

## Links

- Remote endpoint: https://www.gaipagents.com/mcp
- Repository: https://github.com/tjcgraham-rgb/-gaip-agent-to-art-prototype
- Website: https://www.gaipagents.com/
- Changelog RSS feed: https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www.xml
- Changelog JSON feed: https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www.json
- HTML version of this page: https://verifymcp.io/servers/tjcgraham-rgb-gaip-broker/www