1. guides
temper
  • Temper documentation
  • 2026-07-06-cogmap-telos-differentiation
  • handoffs
  • plans
    • archived
    • evidence
  • reviews
  • specs
  • spikes
  • agents
  • auth
    • Temper auth & security
    • The authorization seam (`temper-services::auth`)
    • Cognitive-map & resource authoring authorization
    • JWT verification
    • The issuer / resource-server boundary & the machine-token contract
    • The internal reconcile channel
  • code-reviews
  • cognitive-maps
  • decisions
  • development
  • guides
    • Building a cognitive map from a large corpus
    • Using temper from Claude Desktop (and claude.ai)
    • Cloud Agent Development Guide
    • Ingesting a corpus into a context
    • Development setup
    • Drain operator queries (TraceQL)
    • Enterprise Install — Ground Up
    • Operator guide: read-only GitHub credential via a BYO App + Vercel Connect
    • Operator guide: provisioning a GitHub connection in temper
    • Installing Temper
    • Delivering L0 Kernel Cogmap Content
    • Machine Credentials
    • OpenTelemetry Setup
    • Operational Memory
    • Bootstrapping a Temper org
    • Releasing Temper
    • Self-Hosting Temper with Okta
    • Self-Hosting Temper with a SAML IdP
    • Self-Hosting Temper
    • The Slack Integration — identity, credentials, and what revocation actually means
    • Operator guide: `@temper` on Slack — end-to-end setup
    • Bootstrapping a team's self-cognition cognitive map
    • Working with Teams
    • Integrating Ruby with temper-rb
    • Deploying an Eve agent to Vercel (the steward and the citation auditor)
    • WS6 Collapsed-Schema Dev Environment
  • research
  • security
  • specs
  • doors
    • Building against Temper
    • Running Temper
    • Using Temper
  • Temper /api/query — the compositional read surface
    • Run a declared composition of situated acts.
    • Schemas
      • Composition
      • OutcomeDeclaration
      • ReturnSpec
      • Intention
      • StageNode
      • ActInvocation
      • CombineNode
      • StageInput
      • StageRelation
      • Disclosure
      • QueryResponse
      • StageResult
      • StageOutput
      • ScoreKind
      • Scoring
      • ResourceHit
      • CompositionTrace
      • StageTrace
      • InputSource
      • NarrowedBy
      • StageName
      • ActName
      • IdKind
      • IdSet
      • BoundTerm
      • Extent
      • StageDisposition
      • ActRefusal
      • PlanRefusal
      • RefusalReason
      • ActQuantity
      • QuantityScale
      • ResourceFilter
      • EdgeFilter
      • PropertyPredicate
      • PropertyOp
      • ResourceSection
      • EdgeKind
      • OrdOp
      • CogmapId
      • ContextId
      • IdProvenance
      • FacetPredicate
      • CombineOp
      • BlockId
      • MatchLocation
      • BodyStorage
      • ResourceId
      • IngestState
      • ManagedMeta
      • ProfileId
      • ResourceView
      • Polarity
      • ViaEntry
      • LensId
      • RegionId
      • CogmapRegionRow
      • RegionHit
      • StageInputTrace
      • Standing
      • ActorAuthority
      • Refusal
      • SystemAccessDetails
      • PlanRefusalDetails
      • ErrorDetails
      • ErrorDetail
      • ErrorBody
  • Temper Cloud API
    • Access
      • POST /api/access/requests — submit a join request for the gating team.
      • GET /api/access/requests/me — check own join request status.
      • DELETE /api/access/requests/me — withdraw a pending join request.
      • POST /api/access/reviews — a revoked principal asks an admin to reconsider (spec D15).
      • GET /api/access/settings — read public system settings.
    • Slack Link
      • Disconnect any principal. Operator path — offboarding and stuck users.
      • Disconnect EVERY Slack principal bound to the caller's own profile.
    • Auditor
      • auditor_dispatch
      • auditor_sweep
      • complete_auditor_job
    • Graph
      • POST /api/cogmaps/{id}/graph/slice — R4 cogmap-scoped neighborhood slice.
      • GET /api/graph/cogmaps/{id}/panorama — enter-a-cogmap Tier-0 interior.
      • GET /api/graph/contexts/composition — Beat E Tier-1: the force-graph composition of a container's (or a residual bucket's) members.
      • GET /api/graph/contexts/panorama — Beat E Tier-0: goal-container territories + residual tray.
      • GET /api/graph/home — the you→teams→cogmaps membership home.
      • GET /api/graph/regions/composition — Beat D region→resources composition drill.
    • Cognitive Maps
      • list_cognitive_maps
      • genesis
      • get_cognitive_map
      • reconcile
      • analytics
      • grant_cogmap_access
      • revoke_cogmap_access
      • materialize
      • materialize_delta
      • region_metrics
      • shape
      • bind_team
      • unbind_team
    • Contexts
      • list_contexts
      • create_context
      • get_context
      • context_materialize
      • reassign
      • context_region_metrics
      • rename
      • context_shape
      • share_team
      • unshare_team
    • Events
      • cursor
      • GET /api/graph/elements/{kind}/{id}/trail — R5 element event-trail. kind ∈ {node, edge}.
    • Facets
      • set_facet
      • Read the live facets of one edge.
      • Set a facet whose owner is an **edge** rather than a resource.
      • Read the live facets of one resource — the confirming read for a write that steers region formation and Atlas grouping.
    • Health
      • health_check
    • Ingest
      • create_ingest
      • update_ingest
      • list_blocks
      • append_block
      • finalize_resource
    • Invitations
      • accept
      • decline
      • list_mine
      • list_team_invitations
      • revoke_team_invitation
      • create_team_invitation
    • Invocations
      • list_invocations
      • open
      • show
      • close
    • Profile
      • get_profile
      • update_profile
      • list_auth_links
    • Query
      • `POST /api/query`.
    • Relationships
      • assert
      • fold
      • retype
      • reweight
    • Resources
      • `GET /api/resources` — **one response type, unconditionally.**
      • create_resource
      • get_resource
      • delete_resource
      • update_resource
      • List the finding at `{id}`'s citation-audit trail — one row per audit, each naming its auditor.
      • Record an auditor's signed defensibility verdict on one `(block, source)` citation of the finding at `{id}`. CONFORM to `handlers::edges::assert` (the sibling authored-write handler): thin — build the command, dispatch it, map the error. No persistence he
      • get_content
      • list_resource_edges
      • resource_evidence
      • grant_resource_access
      • revoke_resource_access
      • resource_lineage
      • provenance
      • annotate_resource
    • Meta
      • get_meta
      • update_meta
    • Reassign
      • reassign_resource
      • reassign_team
    • Search
      • `POST /api/search`.
    • Steward
      • candidates
      • dispatch
      • sweep
      • delta
      • advance
    • Teams
      • list_teams
      • create_team
      • detail
      • delete_team
      • update_team
      • add_member
      • remove_member
      • change_role
    • Schemas
      • AcceptInvitationResponse
      • ActInput
      • ActInvocation
      • ActName
      • ActQuantity
      • ActRefusal
      • ActorAuthority
      • AddMemberRequest
      • AdvanceWatermarkAck
      • AdvanceWatermarkRequest
      • AgentAuthorship
      • AppendBlockPayload
      • AssertRelationshipRequest
      • AtlasEdge
      • AtlasHome
      • AtlasNode
      • AtlasSubgraph
      • AuditCitation
      • AuditSweepRow
      • AuditorDispatchTickRequest
      • AuditorDispatchTickResponse
      • AuditorJobCompleteAck
      • BindTeamOutcome
      • BindTeamRequest
      • BlockId
      • BlockProvenanceRow
      • BlocksResponse
      • BodyStorage
      • Bridge
      • BulkReassignAck
      • BulkReassignRequest
      • ChangeRoleRequest
      • CharterBlock
      • CharterDisposition
      • CitationAuditRequest
      • CitationAuditRow
      • ClaimedAuditJob
      • ClaimedJob
      • CloseInvocationRequest
      • CogmapAnalyticsRow
      • CogmapDetail
      • CogmapFoundationRow
      • CogmapGrantBody
      • CogmapId
      • CogmapRegionMetricsRow
      • CogmapRegionRow
      • CogmapRegulationRow
      • CogmapRevokeBody
      • CogmapRow
      • CogmapStaleness
      • CombineNode
      • CombineOp
      • Composition
      • CompositionTrace
      • ConfidenceBand
      • ContentResponse
      • ContextCreateRequest
      • ContextId
      • ContextOwnerRef
      • ContextPanorama
      • ContextRow
      • ContextRowWithCounts
      • CorrelationId
      • CreateCogmapOutcome
      • CreateCogmapRequest
      • CreateInvitationRequest
      • CreateRequestBody
      • CreateReviewBody
      • DeleteResponse
      • DeviceOverrides
      • DispatchTickRequest
      • DispatchTickResponse
      • Disposition
      • DriftSweepRow
      • EdgeFacetRow
      • EdgeFacetSetRequest
      • EdgeFacetsResponse
      • EdgeFilter
      • EdgeId
      • EdgeKind
      • ElementEvent
      • ElementKind
      • EmbedDispatchSummary
      • Entitlements
      • ErrorBody
      • ErrorDetail
      • ErrorDetails
      • EventCursorResponse
      • EventTrail
      • ExactArm
      • ExactHit
      • Extent
      • FacetAck
      • FacetPredicate
      • FacetSetRequest
      • FinalizePayload
      • FoldRelationshipRequest
      • GrantOutcome
      • GraphEdgeRow
      • GroupKeyMeta
      • HealthResponse
      • HomeCogmap
      • HomeContext
      • IdKind
      • IdProvenance
      • IdSet
      • IdpRevocation
      • IngestCreateResponse
      • IngestDelta
      • IngestPayload
      • IngestState
      • InheritedReadGrant
      • InheritedShare
      • InputSource
      • Intention
      • InvitationStatus
      • InvitationTokenRequest
      • InviteeInvitation
      • InvocationAck
      • InvocationActRow
      • InvocationCloseAck
      • InvocationId
      • InvocationSummary
      • InvocationView
      • JoinRequest
      • JoinRequestStatus
      • LensId
      • LineageNode
      • ManagedMeta
      • MatchLocation
      • MaterializeAck
      • MaterializeDelta
      • MaterializeRequest
      • MergePolicy
      • MetaUpdatePayload
      • NarrowedBy
      • NodeHome
      • OpenInvocationRequest
      • OrdOp
      • OrphanNode
      • OutcomeDeclaration
      • PlanRefusal
      • PlanRefusalDetails
      • Polarity
      • Profile
      • ProfileAuthLink
      • ProfileId
      • ProfileUpdateRequest
      • ProfileWithEntitlements
      • PropertyOp
      • PropertyPredicate
      • ProvenanceSource
      • PublicSystemSettings
      • QuantityScale
      • QueryResponse
      • ReassignAck
      • ReassignContextOutcome
      • ReassignContextRequest
      • ReassignResourceRequest
      • ReconcileCogmapRequest
      • ReconcileEdge
      • ReconcileEdgeTombstone
      • ReconcileEntry
      • ReconcileOutcome
      • ReconcileTelos
      • ReconcileTelosBlock
      • ReconcileTombstone
      • Refusal
      • RefusalReason
      • RegionHit
      • RegionId
      • RelationshipAck
      • RemoveMemberOutcome
      • RenameContextOutcome
      • RenameContextRequest
      • ResidualBucket
      • ResidualContext
      • ResidualGroups
      • ResidualOwnedReach
      • ResourceAnnotateRequest
      • ResourceCreateRequest
      • ResourceFacetRow
      • ResourceFacets
      • ResourceFacetsResponse
      • ResourceFilter
      • ResourceGrantBody
      • ResourceHit
      • ResourceId
      • ResourceLineage
      • ResourceListResponse
      • ResourceRevokeBody
      • ResourceSection
      • ResourceSortField
      • ResourceUpdateRequest
      • ResourceView
      • ReturnSpec
      • RetypeRelationshipRequest
      • RevokeOutcome
      • ReweightRelationshipRequest
      • ScoreKind
      • Scoring
      • SearchParams
      • SearchReason
      • SearchResponse
      • SearchResultRow
      • SearchScope
      • SearchScopeInfo
      • SegmentInfo
      • SegmentedBegin
      • SegmentedBeginResponse
      • ShareContextOutcome
      • ShareContextRequest
      • SlackDisconnectRequest
      • SlackDisconnectResponse
      • SlackDisconnectedPrincipal
      • SliceRequest
      • SortOrder
      • StageDisposition
      • StageInput
      • StageInputTrace
      • StageName
      • StageNode
      • StageOutput
      • StageRelation
      • StageResult
      • StageTrace
      • Standing
      • StandingShape
      • Subscription
      • SubscriptionOverride
      • SystemAccessDetails
      • TeamCreateRequest
      • TeamDetail
      • TeamInvitation
      • TeamMemberDetail
      • TeamMemberRow
      • TeamMemberSource
      • TeamRole
      • TeamRow
      • TeamUpdateRequest
      • Territory
      • TerritoryKind
      • TerritoryOverview
      • UnbindTeamOutcome
      • UnshareContextOutcome
      • VaultConfig
      • ViaEntry
      • WideArm
      • WideHit
  1. guides

Using temper from Claude Desktop (and claude.ai)

[observed — 2026-08-01]
Two independent things, in this order. Neither implies the other, and the common failure is doing
only the second and wondering why Claude cannot see anything.
1.
Connect the MCP server — gives Claude access to your knowledge base.
2.
Install the skill — teaches Claude how to use it well.
This guide lives here, not inside the skill bundle, deliberately: it is what you read before you
have the skill, and it is addressed to you rather than to the agent. A copy of it shipped inside the
bundle for a while, where it was both duplicated and read by an agent as if it were instructions.

1. Connect the MCP server#

Settings → Connectors → Add custom connector, with the URL:
https://temperkb.io/mcp
Authentication is OAuth via Auth0 and happens on first connect — Claude will prompt you.
Do not use claude_desktop_config.json for this. That file configures local, stdio MCP
servers. Claude Desktop will not connect to a remote server declared there, so a "url" entry
in mcpServers silently does nothing — it looks configured and never connects. This guide
previously said to do exactly that; it was wrong.
Once connected you get:
SurfaceWhat it is
Resources panel (attachment icon)Browse and attach documents directly. Content is injected before the model runs — no tool call, no tool-call tokens
Toolssearch, list_resources, create_resource, and the rest, called during conversation
Resource URIs are temper://resources/{id}, temper://resources/{id}/content, and
temper://contexts/{ref}/resources — where {ref} is @owner/slug or a UUID. Bare context names
are not accepted.

2. Install the skill#

The connector gives Claude access. The skill gives it the working discipline: which doc types
exist, how goals and tasks are structured, the session-note shape, and the traps (a capped list read
as a complete one; an abbreviated UUID that resolves to nothing).
1.
Download temper-skill-v<version>.zip from the
latest release.
2.
Customize → Skills → +, and upload the zip.
The bundle carries a VERSION file at its root — that is how you tell which one you uploaded, since
the Skills list shows only the skill name.

Verifying what you downloaded#

Each release publishes a .sha256 sidecar beside the zip, and the zip carries a GitHub build-
provenance attestation:
The bundle deliberately carries no per-file manifest, unlike the CLI archives. Manifests exist so
install.sh can verify each extracted file before an atomic swap; nothing installs this bundle, so
there is nothing for one to gate. See
the release guide for the CLI side.

Skills do not sync between surfaces#

A skill uploaded to Claude Desktop / claude.ai is not available in the API or in Claude Code, and
vice versa. Each surface needs its own upload. Claude Code users want the CLI packaging instead
(temper skill install), which is a different tree — it speaks temper … commands rather than tool
calls.

Where things live#

Skill bundle sourceagent-skills/temper-knowledge-base/ (a committed projection — see the generated-artifacts skill)
Build it locallycargo make skill-package → dist/temper-skill-v<version>.zip
Tool referenceknowledge-base.md inside the bundle
Doc types + frontmatterreferences/frontmatter.md inside the bundle (generated from the server's own schemas)
Modified at 2026-08-19 17:59:52
Previous
Building a cognitive map from a large corpus
Next
Cloud Agent Development Guide
Built with