Analytos Brain Omnigraph MCP Server
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Analytos Brain on Omnigraph — Submission POC
This repository is a submission-ready proof-of-concept for the Analytos Org Context Layer / Analytos Brain assessment.
It demonstrates the required loop:
Ingest → Extract → Human Review → Merge to main → Dashboard + MCP → Agents produce real workWhat is included
Omnigraph schema in
omnigraph/schema/analytos.pgOmnigraph query templates in
omnigraph/queries/*.gqCedar policy sketch in
omnigraph/policies/analytos.cedarPython ingestion pipeline that emits Omnigraph-compatible JSONL
Idempotent deterministic IDs for nodes and edges
Review/approval flow with branch diffs and merge attribution
FastAPI dashboard for entity browsing, search, review, and recent changes
MCP wrapper exposing governed graph reads to agents
Content Agent and GTM Agent scripts
Tests covering idempotency, governance, access control, and agent output
Demo seed files matching the requested filenames
Note: The official private seed docs were not available in this chat, so this repo includes realistic demo seed files with the required filenames. If you have the official assessment seed files, replace files in
seed-data/and rerun the same commands.
Related MCP server: provgraf-mcp
Architecture
seed-data/*.md
→ pipeline.ingest
→ Gemini Flash extraction when GEMINI_API_KEY is set; deterministic fallback otherwise
→ runs/<run-id>/graph.jsonl
→ ingest/<run-id> branch
→ human review diff
→ approve merge to main
→ dashboard + MCP wrapper
→ content_agent.py and gtm_agent.pyThe local graph store in .local_graph/state.json is used so the full demo can run without external services. The pipeline also emits Omnigraph-compatible JSONL and includes the real Omnigraph schema/query/policy files for a server-backed deployment.
Local setup
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtOptional MCP dependencies:
npm installOptional Omnigraph install:
curl -fsSL https://raw.githubusercontent.com/ModernRelay/omnigraph/main/scripts/install.sh | bash
omnigraph versionOptional: enable real Gemini LLM extraction
The repo runs without secrets, but for assessment compliance you should set a Gemini key locally or in hosting:
export GEMINI_API_KEY="your-rotated-google-ai-studio-key"
# optional
export GEMINI_MODEL="gemini-1.5-flash"Never commit .env files or paste API keys into the repository.
Run the full local demo
bash scripts/demo.shThis will:
Ingest all seed docs into
ingest/demo-run.Print the diff.
Approve and merge into
mainashuman-reviewer.Run Content Agent.
Run GTM Agent.
Manual workflow
1. Ingest seed data
python -m pipeline.ingest --input seed-data --run-id seed-runExpected result:
branch: ingest/seed-run
status: pending_review
nodes/edges extracted2. Review the branch diff
python -m pipeline.review diff ingest/seed-run3. Approve and merge
python -m pipeline.review approve ingest/seed-runOnly human-reviewer can merge. ingest-agent cannot write directly to main.
4. Start dashboard
uvicorn dashboard.backend.main:app --reload --port 8000Open:
http://localhost:8000Dashboard pages:
/entities— entity browser/search— search approved knowledge/review— pending branch review/recent— merge/commit history
5. Run Content Agent
python agents/content_agent.py "Stockly inventory forecasting"The Content Agent:
Uses only approved main-branch knowledge
Cites graph node IDs
Avoids EmailThread/internal-only nodes
Shows access-control check
6. Run GTM Agent
python agents/gtm_agent.py "Stockly"The GTM Agent returns:
Target company profile
Persona to contact
Example companies
Opening angle grounded in proof points
Graph node citations
MCP usage
Install Node dependencies:
npm installRun content-agent MCP wrapper:
ANALYTOS_ACTOR=content-agent node mcp/custom-wrapper/server.mjsRun GTM-agent MCP wrapper:
ANALYTOS_ACTOR=gtm-agent node mcp/custom-wrapper/server.mjsClaude Desktop-style configs are provided:
mcp/content-agent-config.jsonmcp/gtm-agent-config.json
The MCP tool try_read_email_threads demonstrates policy behavior:
{
"actor": "content-agent",
"visible_count": 0,
"denied_count": 2
}Omnigraph-backed mode
The POC emits JSONL that follows the Omnigraph bulk load shape:
{"type":"Product","id":"product:stockly","data":{...}}
{"edge":"HAS_FEATURE","id":"edge:...","from":"product:stockly","to":"feature:stockly:demand-forecasting","data":{...}}Initialize a real graph:
mkdir -p data
omnigraph init --schema omnigraph/schema/analytos.pg data/analytos-brain.omniLoad an approved run branch into Omnigraph:
omnigraph branch create ingest/seed-run data/analytos-brain.omni
omnigraph load --data runs/seed-run/graph.jsonl --mode merge --branch ingest/seed-run data/analytos-brain.omni
omnigraph branch merge ingest/seed-run --into main data/analytos-brain.omniFor cluster/server deployments, adapt omnigraph/cluster.yaml, then use:
omnigraph cluster validate --config omnigraph/cluster.yaml
omnigraph cluster plan --config omnigraph/cluster.yaml
omnigraph cluster apply --config omnigraph/cluster.yaml
omnigraph-server --cluster omnigraph/cluster.yaml --bind 0.0.0.0:8080Tests
pytest -qCovered criteria:
Idempotent re-ingestion
No direct writes to main by ingest-agent
Merge requires human-reviewer
content-agent cannot read EmailThread nodes
Content Agent has citations and no internal client leak
GTM Agent produces a prospecting brief
Assessment criteria mapping
Criterion | Where implemented |
Governance correctness |
|
Extraction quality |
|
Agent output quality |
|
Access control |
|
Dashboard usability |
|
Engineering hygiene | README, tests, reproducible scripts, clear repo structure |
Known limitations
Gemini Flash extraction is implemented and used when
GEMINI_API_KEYis set. The deterministic fallback remains for reproducible tests and demos without credentials.The local graph store is a test/demo fallback. Production submission hosting should run Omnigraph server with
cluster.yamland the Cedar bundle.The included seed docs are demo fixtures because the private official seed docs were not uploaded here.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
MCP server for querying BrainKB, a knowledge base for neuroscience knowledge graphs.
Public read-only MCP access to digital learning concepts, semantic relationships, and citations.
One governed MCP endpoint for 40+ business systems. Read-only, source-cited answers.
Governed company knowledge over MCP. Synthetic read-only demo; customer Atlas by contract.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceMCP server for the governed knowledge layer, enabling agents to query canonical entries, retrieve conflicts, and manage proposals with strict tenant isolation and audit.-
- AlicenseAqualityDmaintenanceRead-only MCP server exposing a W3C PROV knowledge graph of verified facts with provenance, enabling AI agents to list, search, and check facts while enforcing that writes remain CLI-only.5MIT
- AlicenseNot gradedqualityCmaintenanceEnables MCP-compatible clients to search and query a team knowledge graph, retrieve document context, upload files for indexing, explore entities and connections, and monitor workspace usage and server health.MIT
- FlicenseNot gradedqualityCmaintenanceServes a governed knowledge system of record over MCP, letting agents answer questions with citations drawn only from approved documents and explicitly decline when the record does not cover the question.34 npm-