ADARD: full StarIntel MCP authoritative platform interface #136

Open
opened 2026-08-24 03:11:07 +00:00 by lost-rob0t · 2 comments
lost-rob0t commented 2026-08-24 03:11:07 +00:00 (Migrated from github.com)

Run a canonical ADARD research/design loop for the full StarIntel MCP that will sit beneath the approved Meta MCP. Research the current StarIntel server capability, authorization, scope, API, target/actor, search/document, dataset/view, audit, credential/principal, async/job, and future transport surfaces; map them into a typed MCP interface without bypassing StarIntel authorization or mirroring legacy HTTP routes mechanically. Keep the implementation library-first Common Lisp: generic MCP/LLM tooling remains reusable ASDF systems in the shared monorepo, while the final StarIntel MCP is a StarIntel-specific package consuming those libraries. Define read/write/admin boundaries, resource-vs-tool decisions, schemas, task/stream behavior if warranted, capability discovery, auth propagation, errors, provenance/audit, conformance fixtures, and Meta MCP integration. No infrastructure topology, secrets, deployment, or runtime changes in this ADARD loop. Public research/design should publish through auto-research.starintel.actor; implementation/topology handoff goes only to infrastructure RAGE after approval.

Run a canonical ADARD research/design loop for the full StarIntel MCP that will sit beneath the approved Meta MCP. Research the current StarIntel server capability, authorization, scope, API, target/actor, search/document, dataset/view, audit, credential/principal, async/job, and future transport surfaces; map them into a typed MCP interface without bypassing StarIntel authorization or mirroring legacy HTTP routes mechanically. Keep the implementation library-first Common Lisp: generic MCP/LLM tooling remains reusable ASDF systems in the shared monorepo, while the final StarIntel MCP is a StarIntel-specific package consuming those libraries. Define read/write/admin boundaries, resource-vs-tool decisions, schemas, task/stream behavior if warranted, capability discovery, auth propagation, errors, provenance/audit, conformance fixtures, and Meta MCP integration. No infrastructure topology, secrets, deployment, or runtime changes in this ADARD loop. Public research/design should publish through auto-research.starintel.actor; implementation/topology handoff goes only to infrastructure RAGE after approval.
lost-rob0t commented 2026-08-24 03:17:28 +00:00 (Migrated from github.com)

Operator correction/requirement for PIPELINE-005, 2026-08-23 America/New_York: the implementation phases must explicitly build the full authoritative StarIntel MCP before Meta MCP integrates StarIntel. Required ordering is reusable Common Lisp MCP/LLM libraries -> full StarIntel MCP (with its own ADARD/conformance gate) -> search-provider adapters and StarIntel Meta MCP -> future A2A StarIntel agent. Meta must consume StarIntel through that MCP, never bypass it. Crawl remains excluded from PIPELINE-005.

Operator correction/requirement for PIPELINE-005, 2026-08-23 America/New_York: the implementation phases must explicitly build the full authoritative StarIntel MCP before Meta MCP integrates StarIntel. Required ordering is reusable Common Lisp MCP/LLM libraries -> full StarIntel MCP (with its own ADARD/conformance gate) -> search-provider adapters and StarIntel Meta MCP -> future A2A StarIntel agent. Meta must consume StarIntel through that MCP, never bypass it. Crawl remains excluded from PIPELINE-005.
lost-rob0t commented 2026-08-24 03:26:26 +00:00 (Migrated from github.com)

ADARD review packet (research/design remain REVIEW, not approved):

  • Reuse/extend the existing starintel-gserver-client remote SDK boundary from STAR-RESEARCH-038; do not create a second MCP-specific StarIntel HTTP client.
  • Keep generic MCP protocol/transport/server/client/resource/tool/schema/discovery behavior in reusable Common Lisp ASDF libraries.
  • Add a thin reusable starintel-mcp-adapter in the MCP/LLM-tooling monorepo; final starintel-mcp is a package above that adapter.
  • Use starintel:// resources/templates for addressable reads (identity context, documents, bulk-job status, bounded target collections where accepted).
  • Use tools for bounded computation/mutation: search, semantic view_query, document create/delete/bulk, target dispatch; document update only after its authoritative SDK/auth contract is proven.
  • Discovery = verified StarIntel grants/resource visibility ∩ runtime availability ∩ reviewed MCP exposure policy. Every read/call still reauthorizes independently.
  • Do not advertise current stub/501 lease/replay/event operations.
  • Do not expose bootstrap/login/password or credential create/rotate flows that move raw bearer secrets through model context.
  • Search returns bounded summaries + MCP resource links, not full-document floods.
  • Current process-local bulk jobs return StarIntel job resource handles; do not claim MCP Task semantics until a separate lifecycle gate passes.
  • Align MCP schemas with STAR-RESEARCH-039 machine-readable contract/generated-operation work as it becomes available; never create a third schema authority.
  • PIPELINE-006 read/search conformance is the hard gate before PIPELINE-005 Meta MCP may integrate StarIntel.
  • Crawl remains outside PIPELINE-005. Autonomous research remains future A2A-agent behavior.

Canonical files in PR #135:

  • roam/research/auto-research/STAR-MCP-RESEARCH-001-authoritative-platform-interface.org
  • roam/design/auto-research/STAR-RESEARCH-PIPELINE-006-starintel-mcp-authoritative-platform-interface.org
ADARD review packet (research/design remain REVIEW, not approved): - Reuse/extend the existing `starintel-gserver-client` remote SDK boundary from STAR-RESEARCH-038; do not create a second MCP-specific StarIntel HTTP client. - Keep generic MCP protocol/transport/server/client/resource/tool/schema/discovery behavior in reusable Common Lisp ASDF libraries. - Add a thin reusable `starintel-mcp-adapter` in the MCP/LLM-tooling monorepo; final `starintel-mcp` is a package above that adapter. - Use `starintel://` resources/templates for addressable reads (identity context, documents, bulk-job status, bounded target collections where accepted). - Use tools for bounded computation/mutation: `search`, semantic `view_query`, document create/delete/bulk, target dispatch; document update only after its authoritative SDK/auth contract is proven. - Discovery = verified StarIntel grants/resource visibility ∩ runtime availability ∩ reviewed MCP exposure policy. Every read/call still reauthorizes independently. - Do not advertise current stub/501 lease/replay/event operations. - Do not expose bootstrap/login/password or credential create/rotate flows that move raw bearer secrets through model context. - Search returns bounded summaries + MCP resource links, not full-document floods. - Current process-local bulk jobs return StarIntel job resource handles; do not claim MCP Task semantics until a separate lifecycle gate passes. - Align MCP schemas with STAR-RESEARCH-039 machine-readable contract/generated-operation work as it becomes available; never create a third schema authority. - PIPELINE-006 read/search conformance is the hard gate before PIPELINE-005 Meta MCP may integrate StarIntel. - Crawl remains outside PIPELINE-005. Autonomous research remains future A2A-agent behavior. Canonical files in PR #135: - `roam/research/auto-research/STAR-MCP-RESEARCH-001-authoritative-platform-interface.org` - `roam/design/auto-research/STAR-RESEARCH-PIPELINE-006-starintel-mcp-authoritative-platform-interface.org`
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
nsaspy/starintel-auto-research#136
No description provided.