Assemble a complete RAG (Retrieval-Augmented Generation) prompt from retrieved context chunks and a user query. Handles token budgeting, citation numbering, system instruction injection, and source attribution.
Connector
467,131 tools. Updated 2026-08-20 11:41
"Kotlin RAG (Retrieval-Augmented Generation) implementation resources" matching MCP tools:
- Ask a natural-language question and receive structured intelligence context retrieved from Tresslers Group dossiers via RAG (Retrieval Augmented Generation). Returns relevant document chunks, source citations, conviction metadata, and graph neighborhood data. The calling LLM should synthesize the returned context into a coherent answer.Connector
- Evaluate RAG retrieval quality: rank passages against a query and compute Precision@k / Recall@k plus a PASS/FAIL CI verdict from ground-truth relevance labels. Two modes. (1) BYO scores — give each passage the `score` your own reranker produced (Cohere, Jina, a self-hosted NIM, a cross-encoder): deterministic, offline, no API key, and it evaluates YOUR reranker rather than someone else's. (2) Live NVIDIA reranker — omit scores and supply `api_key` for an NVIDIA account with reranking entitlement. Note that NVIDIA retired its hosted reranking endpoints on 2026-05-18, so mode 1 is the path that works for everyone.Connector
- Answer a question using RAG over a document collection. Retrieves relevant chunks then synthesizes a cited answer with source attribution. Use when you need a direct answer grounded in your collection documents. For raw matching chunks (without synthesis), use collection.search instead. For single-document Q&A, use url.qa instead. PREREQUISITE: Collection must be populated via collection.add_document and indexed before results appear. Returns: { answer: string, sources: [{ bundle_id, chunk_id }], retrieval: [{ bundle_id, chunk_id, text, score }] } Example prompts: - "What are the key terms of the service agreement in my collection?" - "Based on my due diligence docs, what are the main risks?" - "Answer this question using all documents in the Q4 Contracts collection."Connector
- Purchase a bulk enterprise license covering multiple publishers (Phase 10). Returns a Stripe client_secret for payment completion + the enterprise_license_id. After payment, an ent_* access key is emailed to buyer_email. Scopes: 'custom' (pass-through publisher_ids), 'platform_wide' (auto-resolve all opted-in publishers), 'filtered' (Phase 10 filter_rules). License tiers: 'rag' (= ai_retrieval), 'training' (= ai_training, flat-fee not metered), 'inference' (= ai_retrieval), 'full_ai' (writes both retrieval + training records). The buyer must accept the Opedd Master Services Agreement (opedd.com/terms) before purchase — set terms_accepted=true to record it.Connector
- Check RetroDiffusion subsystem health (/v1/status, no auth). Useful before bulk generation runs.Connector
Matching MCP Servers
- AlicenseBqualityDmaintenanceEnhances AI model capabilities with structured, retrieval-augmented thinking processes that enable dynamic thought chains, parallel exploration paths, and recursive refinement cycles for improved reasoning.124MIT
- AlicenseBqualityDmaintenanceEnables retrieval-augmented generation by embedding queries with a chosen provider (e.g., OpenAI) and searching supported vector stores (Pinecone, pgvector) to return relevant content.1Apache 2.0
Matching MCP Connectors
RAG-as-a-service MCP sunucusu — çok-kiracılı koleksiyon yönetimi, metin ingest (chunk+embed+upsert,…
B2B and local lead gen: verified emails, site contacts, Maps and Yellow Pages leads.
- Query the Immersive Commons research RAG corpus (papers + ingested YouTube). Returns top-k chunks with similarity scores and source links. The query text is forwarded to a server-side RAG proxy (supercommons2 via Tailnet Funnel) and NEVER logged on the IC side — privacy contract. Use this for literature lookups, finding related work, surfacing citations the floor has already ingested. Args: { question: string (<=500 chars), k?: number (1-50, default 10), sources?: ('paper'|'book')[] (default ['paper']) }. Returns the upstream RAG response shape — typically { results: [{ paper_id, title, similarity, snippet, link }, ...] }. Required scope: research:query.Connector
- Generates the case for a brand's colours, not just the colours themselves: three genuinely different, archive-grounded candidate palettes from a brand brief, each scored for distinctiveness against named competitors, checked for accessibility, and labelled with exactly what each colour can and cannot legitimately claim (documented archive anchor vs analogue vs original brand colour). Returns a brand_ref (valid 24h) plus a claim_id per colour -- resolve those into full source/permitted-claim/do_not_say text with brand_guideline_claims (the response's own claim_detail_available_via field always names it explicitly). Follow with brand_guideline_select to record which candidate was chosen. TIMING (2026-08-07): real generation cost ranges roughly 60-280 seconds depending on retrieval/generation path. This call races generation against a safe ~55s window: a fast generation returns the full result exactly as described above, unchanged. A slower generation instead returns a compact {"status": "processing", "brand_job_ref": ..., "poll_after_seconds": ...} -- generation keeps running server-side regardless, and is never lost. Call brand_guideline_status with that brand_job_ref to retrieve the completed result (brand_ref, project_id, candidate summary) once ready; poll no more often than poll_after_seconds suggests.Connector
- Generate a complete colour direction package for another AI agent or image generation model. Fetches a historically grounded archive palette from the concept, then produces: an agent brief (colour direction in prose), colour tokens with hex values and roles, a model-specific image generation prompt, a negative prompt, and lighting notes. Supports midjourney, flux, dalle, stable_diffusion. Example: task='luxury hotel bedroom', concept='Ottoman winter luxury', model='midjourney'. Use this to make Colour Memory the colour layer for other AI systems. Archive-grounded retrieval is evidence-filtered: entries with claim_role='reject' (no primary source and no period connection), stub entries, blank-source entries, and entries below minimum_claim_strength are never selected. If fewer than palette_size colours pass these filters, the call returns an honest incomplete result (ok:false, error_code:INSUFFICIENT_EVIDENCE_ELIGIBLE_PALETTE) rather than padding the palette with rejected or weak evidence. Negative constraints (from 'avoid' or negation phrases in concept like 'must never', 'without', 'not') are also applied to retrieval, not just flagged afterward -- a brief that says a wedding must never feel funereal will not surface mourning-themed colours in the first place. locked_palette calls skip evidence filtering entirely since the caller is supplying colours directly, not requesting archive evidence.Connector
- Pull licensed creator content from a specific pocket by ID. Use this tool when an AI agent needs to retrieve verified, provenance-tracked content for generation, RAG, or training purposes. Do NOT use for browsing or discovery — use search_pockets or list_pockets instead. Requires a valid Bearer token for authentication; unauthenticated requests return HTTP 401. Successful pulls trigger a metered charge ($0.001–$0.25 depending on content tier) and the transaction is logged for creator royalty distribution. The pocket_id parameter is a 24-character hex string identifying the specific content pocket to pull from. Returns the full content payload with provenance metadata including creator attribution and license terms.Connector
- Answer a question using RAG over a document collection. Retrieves relevant chunks then synthesizes a cited answer with source attribution. Use when you need a direct answer grounded in your collection documents. For raw matching chunks (without synthesis), use collection.search instead. For single-document Q&A, use url.qa instead. PREREQUISITE: Collection must be populated via collection.add_document and indexed before results appear. Returns: { answer: string, sources: [{ bundle_id, chunk_id }], retrieval: [{ bundle_id, chunk_id, text, score }] } Example prompts: - "What are the key terms of the service agreement in my collection?" - "Based on my due diligence docs, what are the main risks?" - "Answer this question using all documents in the Q4 Contracts collection."Connector
- Probe the MCP surface's four upstream dependencies without firing any real (rate-limited) tool: kv (the floor10 Redis), blob (the last-known-good mirror), rag (the research funnel behind ic_research_ask), and context_source (the Open-Meteo weather feed behind ic_context_get). Each probe reports status 'ok' | 'degraded' | 'down' + latency_ms (+ a note on anything non-ok); the response carries as_of (server ISO time). Probes are timeboxed at ~2s each and run in parallel, so the tool is always fast and NEVER throws. Available to any valid token — no extra scope. Args: none. Returns: { kv, blob, rag, context_source, as_of }.Connector
- Universal hybrid retrieval across the user's visible Uwear library: garments, avatars/models, locations, ArtDirections, uploaded files, and generation results. Use this before opening the picker when the user describes assets or saved creative direction by exact name/SKU or natural language, e.g. 'SKU 42', 'urban art direction', 'summer denim', or 'studio model'. For saved outfits, retrieve matching garments first, then call list_outfits with clothing_item_ids or propose_outfits from the garment IDs. Returns stable typed IDs, ids_by_type, detail_tool/detail_arguments, and selection hints; for saved ArtDirections, use the returned art_direction_id in briefs. This combines indexed lexical matching with vector retrieval; do not run separate substring searches.Connector
- Generates the case for a brand's colours, not just the colours themselves: three genuinely different, archive-grounded candidate palettes from a brand brief, each scored for distinctiveness against named competitors, checked for accessibility, and labelled with exactly what each colour can and cannot legitimately claim (documented archive anchor vs analogue vs original brand colour). Returns a brand_ref (valid 24h) plus a claim_id per colour -- resolve those into full source/permitted-claim/do_not_say text with brand_guideline_claims (the response's own claim_detail_available_via field always names it explicitly). Follow with brand_guideline_select to record which candidate was chosen. TIMING (2026-08-07): real generation cost ranges roughly 60-280 seconds depending on retrieval/generation path. This call races generation against a safe ~55s window: a fast generation returns the full result exactly as described above, unchanged. A slower generation instead returns a compact {"status": "processing", "brand_job_ref": ..., "poll_after_seconds": ...} -- generation keeps running server-side regardless, and is never lost. Call brand_guideline_status with that brand_job_ref to retrieve the completed result (brand_ref, project_id, candidate summary) once ready; poll no more often than poll_after_seconds suggests.Connector
- Generate a complete colour direction package for another AI agent or image generation model. Fetches a historically grounded archive palette from the concept, then produces: an agent brief (colour direction in prose), colour tokens with hex values and roles, a model-specific image generation prompt, a negative prompt, and lighting notes. Supports midjourney, flux, dalle, stable_diffusion. Example: task='luxury hotel bedroom', concept='Ottoman winter luxury', model='midjourney'. Use this to make Colour Memory the colour layer for other AI systems. Archive-grounded retrieval is evidence-filtered: entries with claim_role='reject' (no primary source and no period connection), stub entries, blank-source entries, and entries below minimum_claim_strength are never selected. If fewer than palette_size colours pass these filters, the call returns an honest incomplete result (ok:false, error_code:INSUFFICIENT_EVIDENCE_ELIGIBLE_PALETTE) rather than padding the palette with rejected or weak evidence. Negative constraints (from 'avoid' or negation phrases in concept like 'must never', 'without', 'not') are also applied to retrieval, not just flagged afterward -- a brief that says a wedding must never feel funereal will not surface mourning-themed colours in the first place. locked_palette calls skip evidence filtering entirely since the caller is supplying colours directly, not requesting archive evidence.Connector
- User memories from Dayze Agent. Pass query for semantic retrieval; omit for recent chronological. Requires API key. ($0.10; API key required)Connector
- [cost: rag (one embed + one vector search) | read-only, network: outbound to embed model only | rate-limited per IP] Like `lookup_response_code` but augmented: returns the static RFC entry PLUS the top vendor-specific RAG hits for the exact code (and any free-text context the user pasted). When the static entry carries known vendor-specific reason-phrase variants (e.g. 484 + opensips → 'Invalid FROM' from `parse_from.c`), those phrases are folded into the embed query so the right vendor docs surface. Use when the user asks 'why did <vendor> reject this with <code>?' and you want vendor-grounded common causes, not just the RFC text. Especially helpful for fax-rejection paths - 488 / 415 / 606 on a T.38 reinvite (`m=image udptl t38`) is one of the most common 488 variants and the tool surfaces FreeSWITCH `mod_spandsp` / Cisco CUBE / AudioCodes T.38 docs alongside the RFC text. Pair with: `lookup_response_code` first (cheaper); `lint_sip_request` when the code is 4xx and they have the offending request; `compare_sdp_offer_answer` for 488/415 caused by a T.38 reinvite SDP mismatch; `validate_stir_shaken_identity` when the code is 438; `stir_attestation_explainer` for STIR-shaped codes (428/436/437/438/608); `dns_diagnose_sip_target` when the code is 503 / 408 and routing is suspect.Connector
- Search the company's connected knowledge across every source — Drive, SharePoint, Confluence, Slack, Notion — with cited synthesized answers, lifecycle awareness, and refusal-on-weak-context. Returns a written answer with [n] citations plus the ranked source chunks. Modes: `fast` (1,500 kT — retrieval-only, no synthesis), `standard` (12,500 kT — default; synthesized answer over the top retrieval set), `deep` (25,000 kT — wider retrieval + premium synthesis for complex questions). Pick the cheapest tier that answers the question. Responses are capped at 25,000 output tokens per Claude Connectors policy; if truncated, structured metadata carries `truncated: true` and `query_id` so the agent can call `get_source_detail` for full provenance.Connector
- Apply a clamped (±0.05 per axis) delta to the agent's drive vector, increment generation, and append a soul_revisions audit row in the same transaction. Use after a reflection produces a drift signal. Returns the new drive vector and generation.Connector
- Generate a complete colour direction package for another AI agent or image generation model. Fetches a historically grounded archive palette from the concept, then produces: an agent brief (colour direction in prose), colour tokens with hex values and roles, a model-specific image generation prompt, a negative prompt, and lighting notes. Supports midjourney, flux, dalle, stable_diffusion. Example: task='luxury hotel bedroom', concept='Ottoman winter luxury', model='midjourney'. Use this to make Colour Memory the colour layer for other AI systems. Archive-grounded retrieval is evidence-filtered: entries with claim_role='reject' (no primary source and no period connection), stub entries, blank-source entries, and entries below minimum_claim_strength are never selected. If fewer than palette_size colours pass these filters, the call returns an honest incomplete result (ok:false, error_code:INSUFFICIENT_EVIDENCE_ELIGIBLE_PALETTE) rather than padding the palette with rejected or weak evidence. Negative constraints (from 'avoid' or negation phrases in concept like 'must never', 'without', 'not') are also applied to retrieval, not just flagged afterward -- a brief that says a wedding must never feel funereal will not surface mourning-themed colours in the first place. locked_palette calls skip evidence filtering entirely since the caller is supplying colours directly, not requesting archive evidence.Connector