oprow-continuum
MCP server that acts as a temporal verification firewall, checking structured assumptions or diffs against immutable version receipts to return allow/block/abstain with cited evidence.
README
Oprow Continuum
Oprow Continuum is a temporal verification firewall for coding agents. Before
an agent changes a repository, Oprow checks its structured assumptions or
unified diff against immutable version receipts and returns allow, block,
or abstain with exact evidence.

The product contract is deliberately stricter than ordinary retrieval:
- CockroachDB stores immutable receipts and supersession edges;
- normal review activates a bounded evidence packet;
- contradicted assumptions block;
- unsupported semantics abstain rather than receive approval;
- review never mutates the repository;
- state-changing actions remain separately confirmation-gated.
Status
The submitted product contract is
docs/AGENT_VERIFICATION_ENDGAME_2026-08-14.md.
The concise calibration record is
docs/EVIDENCE_LEDGER.md; closed plans remain in
docs/archive/.
This repository contains a runnable trusted-core application, agent API, MCP server, and operator UI. It does not claim product superiority or bundle a trained semantic checkpoint. The service exposes explicit version semantics, exact cited receipt reads, confirmation-gated rollback proposals, stale-proposal rejection, fixed active state accounting, and checkpoint/restore. The CockroachDB immutable point-read adapter passed its first live Cloud gate on 2026-08-05.
The tested learned Shared-Stream Modus_X controller is excluded from the product after the valid A2C/A2D failures. The system does not claim to compress an entire codebase into one small controller or understand arbitrary code semantics.
Product Target
repository history -> typed extraction -> CockroachDB temporal receipts
agent claim or diff -> bounded verification -> allow / block / abstain
-> cited evidence -> human decision
Judges can connect a bounded public GitHub repository directly from the UI. A self-hosted operator can also exercise the trusted core with a pinned event stream at startup:
python -m pip install -e ".[product,mcp]"
$env:OPROW_REASONER="disabled"
$env:OPROW_API_TOKEN="replace-with-a-long-random-value"
oprow serve --repository-id fastapi/fastapi `
--events examples/fastapi_compatibility_events.jsonl
The bounded public demo connector accepts public GitHub repositories under strict import-count, commit, event, checkout-size, and wall-time limits. Private repositories remain unsupported.
Hackathon
Oprow Continuum is registered for the CockroachDB x AWS Agentic Memory Hackathon. CockroachDB will hold the transactional event ledger, capsule metadata, and provenance. Oprow remains the fixed-size active semantic memory. AWS provides CodeBuild/ECR packaging, Secrets Manager runtime credentials, Bedrock reasoning, and the public HTTPS application environment.
See docs/HACKATHON.md.
The component boundary is defined in docs/PRODUCT_ARCHITECTURE.md, with calibrated claims and failures in docs/EVIDENCE_LEDGER.md.
Development
python -m pip install -e ".[dev,product,mcp]"
pytest
oprow demo
oprow db-smoke --env-file .env
oprow serve --repository-id fastapi/fastapi --events examples/fastapi_compatibility_events.jsonl
$env:OPROW_BASE_URL="http://127.0.0.1:8000"
oprow-mcp
powershell -ExecutionPolicy Bypass -File scripts/check_scope.ps1
FastAPI publishes the machine contract at /openapi.json; the review endpoint
is POST /api/agent/review. oprow-mcp exposes the same contract as four MCP
tools over stdio for Claude Code, Cursor, Codex, and other MCP clients.
The chat surface is available at http://127.0.0.1:8000. Without a
database URL it uses a process-local ledger. With DATABASE_URL or
COCKROACH_DB_URL, it uses the tested CockroachDB immutable receipt adapter and
automatically writes .oprow/continuum.capsule after accepted state changes.
The default reasoner is Amazon Nova 2 Lite through Bedrock. Set
OPROW_REASONER=disabled only for trusted-core diagnostics.
The deployed demo uses three pinned, non-synthetic FastAPI compatibility
migrations: dropping Python 3.9, admitting Starlette 1.0+, and raising the
Pydantic minimum. scripts/build_fastapi_compatibility_demo.py reproduces the
six events from the official Git history; the pinned commits and values are
recorded in examples/fastapi_compatibility_manifest.json.
Container smoke:
docker build -t oprow-continuum .
docker run --rm -p 8000:8000 -e OPROW_API_TOKEN=change-me oprow-continuum
Non-loopback binds are rejected unless OPROW_API_TOKEN is set. The operator
enters that token with the key button; it is retained only in browser session
storage and sent on state-changing requests.
Judge Flow
- Connect a supported public GitHub repository or use the pinned FastAPI demo.
- Paste an agent-generated unified diff containing a repository assumption.
- Observe
blockfor contradicted history,allowfor verified current assumptions, orabstainwhen Oprow lacks sufficient evidence. - Inspect immutable CockroachDB receipt citations and the stable review ID.
- Repeat through
oprow-mcpto use the same contract from a coding agent.
The review path never edits the repository. Durable history can grow in CockroachDB while the active hot receipt cache remains bounded. This is not a claim of infinite information capacity or universal code understanding.
License
MIT. See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
E2B
Using MCP to run code via e2b.