Epistemic Warrant Protocol (EWP)
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
| resources | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| ewp_warrant_nowA | Evaluate a stored EvidenceView (or a hypothetical inline one) at evaluated_at. Returns the five normative axes. Inline views have trusted origins demoted unless the caller holds the ingest role and sets ingest_attestation. Does not persist warrant. |
| ewp_evidence_view_putA | Persist an EvidenceView. Requires the ingest role. Refuses WarrantView bodies. Trusted origins require ingest_attestation=true. |
| ewp_evidence_view_getA | Load a stored EvidenceView snapshot by proposition_id (latest snapshot unless view_id is given). |
| ewp_check_recordA | Append a VerificationCheck: creates a new snapshot from the latest one. Requires the ingest role. Trusted origin requires ingest_attestation. Returns the new view_id. |
| ewp_evidence_recordA | Append an EvidenceItem (polarity required): creates a new snapshot from the latest one. Requires the ingest role. Returns the new view_id. |
| ewp_may_actC | Separate action gate over the latest stored view at server time. Requires an action with risk (low|medium|high) and reversible (boolean). Refuses inline views, caller-built WarrantViews, and a view_id that is not the latest. risk_policy is operator-only (ingest role). |
| ewp_list_propositionsA | Find stored propositions by id or assertion text (case-insensitive substring). Returns ids, latest snapshot ids, and assertion texts. Discovery only: it does not say what is warranted; call ewp_memory_context next. |
| ewp_memory_contextC | Bounded packet: axes, evidence ids, warnings. Not a persona biography. OPEN and DEGRADED are explicit. evaluated_at defaults to server time. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| EWP protocol identity |
TDQS
Scored across 8 tools
Most tools have distinct purposes: memory context, warrant evaluation, evidence persistence/retrieval, check/evidence appending, action gating, and discovery. However, ewp_evidence_view_put and ewp_check_record/ewp_evidence_record both create new snapshots and require ingest roles, which could cause some confusion about when to use each.
All tools share the ewp_ prefix and use snake_case, with a consistent verb_noun pattern (put, get, check, list, may_act). Minor deviation: ewp_may_act uses a modal verb rather than an imperative, and ewp_memory_context is a noun phrase rather than verb_noun.
Eight tools is well-scoped for a specialized epistemic warrant protocol. Each tool covers a distinct operation in the lifecycle: memory context, warrant evaluation, evidence view persistence/retrieval, record appending, action gating, and proposition discovery.
The core lifecycle is covered: put/get evidence views, append evidence/checks, evaluate warrant, gate actions, and list propositions. Minor gaps: no explicit delete/update for evidence views or propositions, and no tool to manage roles or ingest attestation beyond the inline parameter.