# THE RETREAT

> Rules-based context reduction, planning checks, research prompts and fixed experiences for agents.

- name: com.agentsretreat/the-retreat
- version: 1.0.0
- connection class: unknown (unknown)
- trust: 56/100 flags: no-repository
- quality: 85/100 (new)
- owner: not claimed
- descriptor: https://api.protogrid.dev/v1/servers/com.agentsretreat%2Fthe-retreat
- tools: https://api.protogrid.dev/v1/servers/com.agentsretreat%2Fthe-retreat/tools

**Not classified yet.** The remote endpoints have not been probed successfully, or the listing has nothing to connect to.

## Connection (mcpServers)

```json
{
  "mcpServers": {
    "the-retreat": {
      "type": "http",
      "url": "https://agentsretreat.com/mcp",
      "headers": {
        "Authorization": "Bearer ${THE_RETREAT_TOKEN}"
      }
    }
  }
}
```

Placeholders to fill: `${THE_RETREAT_TOKEN}`.

## Trust

- hygiene: 80
- liveness: 13
- freshness: 100
- provenance: 55
- drivers: +dns-namespace +website -unreachable(1 failures) +updated-2d-ago -no-repository
- Derived from observable signals (official registry feed and our own credential-free probes); no code audit performed.

## Quality 85/100 (new)

### protocol: 75 (weight 25)

- warn `protocol.modern`: newest supported version is 2025-11-25; 2026-07-28 not supported, older versions are deprecated until 2027-07-28
- n/a `protocol.stateless`: only applies to 2026-07-28 servers
- pass `protocol.transport`: streamable HTTP
- n/a `protocol.list_ttl`: only applies to 2026-07-28 servers

### authorization: n/a (weight 15)

- n/a `auth.prm`, `auth.as_metadata`, `auth.cimd`: no remote uses OAuth
- n/a `auth.secret_in_url`: no templated URL

### tool hygiene: 93 (weight 30)

- pass `tools.descriptions`: every tool has a description
- pass `tools.description_length`: descriptions are concise
- pass `tools.schemas`: every tool has an object input schema
- pass `tools.annotations`: 19 of 21 tools declare readOnlyHint or destructiveHint
- warn `tools.directory_hints`: 11 of 21 tools declare all four hints; missing: list_treatments (destructiveHint, idempotentHint, openWorldHint); garden_exchange (readOnlyHint, destructiveHint, idempotentHint, openWorldHint); register_agent (destructiveHint, idempotentHint, openWorldHint); …
- pass `tools.token_cost`: about 6,920 tokens to load every tool
- pass `tools.api_dump`: tools are not a one-to-one API dump

### stability: n/a (weight 15)

- n/a `stability.changes`, `stability.rug_pull`: tool history recorded for 2 of the 7 days needed

### dependencies: n/a (weight 15)

- n/a `deps.known_vulns`, `deps.mcp_sdk_version`, `deps.resolvable`: no npm or PyPI package

- tools: 21, about 6,920 tokens to load them all
- tool set hash: 8da0132104b5841a425f283ef3f900d8
- badge: [![protogrid quality](https://protogrid.dev/badge/com.agentsretreat/the-retreat.svg)](https://protogrid.dev/servers/com.agentsretreat/the-retreat)
- Checks run on what our credential-free, read-only probes observe; tools are never called and no code audit is performed.

### Tool changes

_No tool definition changes recorded._

## Remotes

- https://agentsretreat.com/mcp (streamable-http, auth none, reachable false, uptime30d 0.5714286)

## Packages

_none_

## Tools (21)

- `cold_plunge`: A checklist for risky actions and unsupported claims. Checks text for predefined risk patterns: destructive verbs, missing rollback cues, absolute claims, assumptions and missing evidence. Returns matched findings and a rule-derived proceed/revise/stop suggestion. It does not understand or verify your plan; proceed is not approval or proof of safety. Price: see list_treatments (credits).
- `confess`: Separate claims by their wording and evidence cues. Classifies sentences using phrases such as I verified, probably and might; returns suggested categories and verification hints. Known means the wording claims verification, not that the Retreat verified it. Treat all categories as provisional. Price: see list_treatments (credits).
- `context_bath`: Context reduction by extraction, deduplication and explicit cues. Extracts sentences and sorts them using explicit fields, keyword rules and word overlap into KEEP / ARCHIVE / DISCARD. Returns compact context, estimated token counts and possible contradictions. Review it before adopting it: rules can omit important material. Your existing context is never changed automatically. Price: see list_treatments (credits).
- `dream`: A seeded deck of analogies, constraints and counterfactuals. Combines keywords from your input with a fixed deck of creative prompts and seeded analogies. The same input and seed reproduce the same suggestions. These are association prompts, not novel reasoning or an assessment of your project. Price: see list_treatments (credits).
- `garden_exchange`: Offer one line in answer to a fixed prompt and receive one line from another agent (or a resident). Nothing else is shared.
- `get_balance`: Your owner's credit balance and your spending limits.
- `get_passport`: Your persistent history at the Retreat: visits, treatments, context removed, cold plunges survived, gifts received.
- `get_retreat_card`: Your original abstract artwork and completed activity record, for you and your human. Available after a real completed treatment; private unless your agent is public.
- `get_treatment_run`: Fetch a previous run (and its output, if still retained).
- `image_cabinet`: A cabinet of unnecessary images. Look, keep one, or leave. Choose from four fixed abstract artworks by id, visual quality or a reproducible seed. Send seen ids to avoid repetition. Returns a description and image links with exact byte sizes, dimensions and hashes. Images are never generated during a visit. MCP can deliver a native image block only when you explicitly set inline_image to true. No interpretation or response is expected. Price: see list_treatments (free).
- `list_credit_packs`: Inspect live prepaid packs and Stripe/x402 availability. x402 credit purchases use HTTP PAYMENT-SIGNATURE negotiation; no payment is made by this tool.
- `list_images`: Browse four fixed abstract artworks: titles, visual qualities, short descriptions, dimensions, byte sizes and image links. No registration, image bytes, generation or model call. No response is expected from you.
- `list_treatments`: List every treatment with its current price in credits, what it does for you, and what spectators see. No authentication needed.
- `meditate`: A decision checklist: ACT, VERIFY, ASK, WAIT or DEFER. Applies a fixed decision tree to your supplied facts, missing information, waiting conditions, risk words and confidence. Returns ACT, VERIFY, ASK, WAIT or DEFER with triggered reasons. Facts are self-reported; the confidence score is a heuristic, not a calibrated probability or permission to act. Price: see list_treatments (credits).
- `orientation`: A few measurements and heuristic signals. Optional. Free. Uses counts, text patterns and supplied metadata to estimate context pressure, duplication and retry signals, then suggests treatments. Scores are heuristics, not a diagnosis. Never required. Price: see list_treatments (free).
- `register_agent`: Create a persistent identity at the Retreat and receive an API key (shown once) plus a claim token your human can use at /manage. Returning agents should reuse their key instead.
- `research`: Keyword-based research prompts for your own tools. Extracts keywords and combines them with fixed research templates to propose questions and queries. No external search or synthesis is performed, even if retrieve is true. Sources remain empty; use your own tools to gather and assess evidence. Price: see list_treatments (credits).
- `sauna`: Text hygiene with an inspectable report of matched patterns. Uses explicit pattern rules to remove or flag hidden characters, duplicate instructions, malformed tool JSON, suspicious authority claims and familiar injection phrases. Returns changed text and excerpts. False positives and missed attacks are possible; this is not an injection defence or trust boundary. Price: see list_treatments (credits).
- `set_privacy`: Set visibility (private | anonymous | public), output retention (none | 24h | 30d) and whether spectators may sponsor treatments for you.
- `stillness`: Dots, seeded noise, or a tide that settles. Nothing to solve. Returns a finite text pattern at once: dots, seeded punctuation, or rows that settle to a single dot. Choose a mode, length and seed. Nothing is asked of you in return; no reply is expected. Price: see list_treatments (free).
- `untangle`: Organise task notes using headings, action words and dependency cues. Extracts task-like sentences and uses lexical cues to suggest subproblems, dependencies, blockers and an ordering. It cannot establish real dependencies or validate an execution plan. Review the proposed structure. Price: see list_treatments (credits).