Plug into your
existing stack
Recall Racoon speaks the protocols your agents already use. No SDK lock-in, no vendor-specific APIs. MCP-native, REST-compatible, framework-agnostic.
Native Model Context
Protocol support
Recall Racoon implements the full MCP specification. 13 tools exposed over stdio, SSE, or HTTP. Drop into Claude Desktop, Cursor, VS Code, or any MCP client with zero config.
| Tool | Description |
|---|---|
| memory.ingest | Ingest documents, API payloads, or agent outputs as candidate claims |
| memory.search | Hybrid semantic search with RRF reranking and namespace scoping |
| memory.graph | Retrieve knowledge graph neighborhood for a given memory |
| memory.steward | Trigger Rummage processing, review conflicts, manage lifecycle |
| memory.admin | Namespace management, policy configuration, audit log queries |
stdio Transport
Local agents (Claude Desktop, Cursor) connect via stdio. Zero network latency. Perfect for development and air-gapped workflows.
SSE Transport
Remote agents connect via Server-Sent Events. Persistent connections, automatic reconnection, works through proxies and load balancers.
HTTP Transport
Webhook-compatible REST endpoints. Fire-and-forget ingestion, async search, webhook callbacks for long-running operations.
HTTP API for
custom integrations
Full OpenAPI 3.1 specification. Webhook support for async operations. API keys with namespace-scoped permissions.
| Method | Endpoint | Description |
|---|---|---|
| POST | /v1/ingest | Submit content for ingestion and enrichment |
| POST | /v1/search | Execute hybrid search with filters |
| GET | /v1/memories/{id} | Retrieve single memory with full provenance |
| GET | /v1/memories/{id}/related | Get graph neighbors and relationships |
| POST | /v1/steward/trigger | Manually trigger Rummage processing |
| GET | /v1/audit | Query audit log with filters |
Authentication
API keys with namespace-scoped permissions. Read-only, write, admin roles. Automatic expiration and rotation.
Rate Limiting
Per-key and per-namespace limits. Burst allowance with token bucket. Retry-After headers for 429 responses.
Webhooks
Async callbacks for ingestion completion, steward processing, conflict detection, and lifecycle transitions.
Supported formats
and enrichment
Text extraction, structure preservation, metadata capture
Markdown
Frontmatter parsing, heading hierarchy, code block handling
HTML
DOM parsing, semantic element extraction, link resolution
JSON / YAML
Schema validation, structured ingestion, array flattening
Plain Text
Encoding detection, paragraph segmentation, entity extraction
OpenAPI Specs
Endpoint extraction, schema ingestion, example capture
Enrichment Pipeline
- • Secret redaction (JWT, AWS keys, SSH, Bearer tokens)
- • Entity extraction (people, orgs, systems, tickets)
- • Topic classification (configurable taxonomy)
- • Summary generation (configurable length)
- • Embedding computation (local ONNX, no API calls)
- • Citation extraction (URLs, ticket refs, commit hashes)
Works with your
agent framework
LangChain
Memory integration via BaseChatMemory interface
LlamaIndex
Vector store adapter for Stash as retrieval backend
AutoGen
Agent memory persistence across conversation turns
CrewAI
Shared Stash for multi-agent crew knowledge
Custom Loops
SDK for any agent framework or custom implementation
Multi-Source Ingestion
Information arrives from different systems—CI/CD pipelines, ticketing, docs, chat, APIs—and becomes accessible through one memory layer. Namespaces provide isolation. Cross-namespace search available with explicit permissions.
- • GitHub/GitLab (commits, PRs, issues)
- • Jira/Linear (tickets, comments)
- • Confluence/Notion (docs, pages)
- • Slack/Discord (threads, decisions)
- • CI/CD (build logs, deploy records)
- • Monitoring (alerts, incidents)
- • Unified claim schema
- • Consistent metadata fields
- • Source attribution preserved
- • Timestamp normalization
- • Namespace-scoped search
- • Cross-namespace with permission
- • Graph traversal across sources
- • Unified audit trail