How battam1111/omniseek works
OmniSeek is a self-hosted MCP server that gives an agent a unified way to search and inspect information across web, research, social, and login-walled sources. The main workflow is an MCP client invoking server tools; search fans out through source adapters, then normalizes and ranks results, while read, view, and transcribe handle documents and media. Research tools add identity, citation, enrichment, and collaboration analysis; recall adds persistent text/vector retrieval and an evidence graph, while sensors support standing queries. README tool descriptions and the source index establish the broad subsystems; some internal wiring is unsampled, so uncertain links are omitted.
Main entry points: server.py MCP tools · fetcher.py Source router · api API adapters · walled Login-walled adapters
Components 22
MCP interface
- MCP tools
src/omniseek/server.py
Search and sources
- Source router
src/omniseek/core/fetcher.py - API adapters
src/omniseek/core/sources/api - Scrape adapters
src/omniseek/core/sources/scrape - Login-walled adapters
src/omniseek/core/sources/walled - Document normalization
src/omniseek/core/normalize.py - Deduplicate and rank
src/omniseek/core/rank.py
Media and research
- Document reader
src/omniseek/core/docreader.py - Local transcription
src/omniseek/core/asr.py - Image and video view
src/omniseek/core/vframes.py - Research analysis
src/omniseek/core/cartographer.py - Paper enrichment
src/omniseek/core/enrich.py - Identity and relations
src/omniseek/core/relations.py
Recall and monitoring
- Recall search
src/omniseek/core/recall/store.py - Evidence graph
src/omniseek/core/recall/graph.py - Recall ingestion
src/omniseek/core/recall/writer.py - Standing queries
src/omniseek/core/sensor.py
Runtime services
- Response cache
src/omniseek/core/cache.py - HTTP and upstream controls
src/omniseek/core/http.py - Credential loader
src/omniseek/core/auth.py
Other components
- Agent / client
- External sources
Connections 24
- Agent / client → MCP tools: invokes tools
README.md - MCP tools → Source router: searches sources
src/omniseek/server.py - Source router → API adapters: dispatches queries
src/omniseek/core/sources/api/openrouter_rankings_source.py - Source router → Scrape adapters: dispatches queries
src/omniseek/core/fetcher.py - Source router → Login-walled adapters: dispatches queries
src/omniseek/core/sources/walled/xiaohongshu_cn_source.py - API adapters → External sources: requests data
- Scrape adapters → External sources: requests data
- Login-walled adapters → External sources: requests data
- Source router → Document normalization: normalizes results
src/omniseek/core/fetcher.py - Source router → Deduplicate and rank: ranks results
src/omniseek/core/fetcher.py - MCP tools → Document reader: reads documents
README.md - MCP tools → Local transcription: transcribes media
README.md - MCP tools → Image and video view: views media
README.md - MCP tools → Research analysis: maps citations
README.md - MCP tools → Paper enrichment: enriches papers
README.md - MCP tools → Identity and relations: resolves researchers
README.md - MCP tools → Recall search: recalls documents
README.md - MCP tools → Evidence graph: queries graph
README.md - Recall ingestion → Recall search: writes index
src/omniseek/core/recall/writer.py - MCP tools → Standing queries: runs monitoring
src/omniseek/server.py - Source router → Response cache: reads and writes
src/omniseek/core/fetcher.py - Source router → HTTP and upstream controls: makes requests
- API adapters → Credential loader: loads credentials
src/omniseek/core/sources/api/openreview_source.py - Login-walled adapters → Credential loader: loads credentials