Best for
- Teams needing self-hosted AI platform
- Enterprises requiring RAG and advanced search
- Developers wanting custom AI agents
O
Onyx
Community
Not a standard server
Open Source AI Platform - AI Chat with advanced features that works with every LLM
This is not a typical connect-and-use MCP server; read the source repository before using it.
- Category
- Other #222 of 230
- Stars
- ★ 32.4k Very popular
- Transport
- stdio (local process)
- Runtime
- Docker · Python
- Credentials
- Optional API key
- License
- Other / unspecified
- Last commit
27FMRS · D
Onyx is a powerful open-source AI platform suitable for enterprises and developers needing full AI features, but deployment and maintenance complexity is high.
Strongest · Maintenance 13/20 Weakest · Reliability 1/20
Why each score
- Reliability 1/20
- Core finding: this repository is a full AI application platform (RAG, chat UI, 50+ connectors, background indexing workers), not an MCP server. The review target's root contains no MCP server manifest (SERVER MANIFEST not cached) and no tool declarations, so the MCP init + tool-list handshake cannot be performed. The README only describes MCP as a way for Onyx agents to interact with external applications, i.e. Onyx acts as an MCP client. No CI or committed tests covering a server key path were supplied. The static calibration cap is 12, and the actual evidence is far below it; the contradiction between the 'MCP server' framing and the actual standalone application makes the score near zero.
- Security and permissions 5/20
- No red-line evidence (malware, credential theft, covert exfiltration, real tokens in install examples) is present in the supplied material, so the 0-4 red-line band is not triggered. However, credential handling, least privilege, and data boundaries are unverifiable because no source code was supplied; per the evidence rules, missing evidence is a deduction, not the best case. Visible concerns: the README promotes a curl | bash one-line install (executes an unvetted remote script, typically with elevated privileges) and the platform's surface is broad (web search, code execution, custom code, many indexing connectors), implying wide data access. Positive signals (sandboxed code execution, SSO via OAuth/OIDC/SAML, RBAC, audit history) exist only as README claims and cannot be verified, so a conservative mid-low score is given.
- Maintenance 13/20
- The repository is not archived, is extremely popular (31k+ stars), references a v3.0.0 release asset and a Feb 2026 benchmark claim, and provides Discord, CONTRIBUTING.md, and documentation entry points, indicating an active project with community governance. Deductions: GitHub license metadata is NOASSERTION while the README claims MIT for the CE edition (plus a separate EE edition), leaving licensing governance ambiguous; no security policy or security-response channel is visible; 392 open issues suggest some backlog; the supplied material contains no commit history, release cadence, or dependency-update evidence to verify maintenance frequency.
- Documentation 6/20
- The README is solid at the application level: feature lists, Lite vs Standard deployment modes, cloud/Helm/Terraform guidance, enterprise capabilities, licensing, and community links are reasonably layered. But as an MCP server, documentation is entirely absent: there are no tool parameters, no auth flow, no limits/cost disclosure, no troubleshooting guide, and no client-connection examples. The external docs site (docs.onyx.app) is not part of the supplied source and cannot serve as reviewable evidence, and README self-claims such as 'best in class' and 'top of leaderboard' are unverifiable in the files. Hence only a mid-low score based on the application-level docs.
- Setup experience 2/20
- From an application standpoint, the README advertises a one-command install and a Lite mode under 1GB memory, which simplifies deployment; however, this is not an MCP-server installation path. As an MCP server, the steps from preparing a runtime to a working connection in a mainstream client are missing entirely: the repository root has no MCP server entry point, no manifest, and no client configuration examples. Standard mode requires Docker, Redis, MinIO, background workers, and model-inference containers — a complex stack — and none of the deployment claims can be verified with execution evidence from the supplied source. Static calibration caps setup at 15, and here even the basic path does not exist, so the score is minimal.
Static review · not runListed 2026-08-07
Read the FMRS scoring method →Fit and risk
What it can accessReads local filesRuns commands or codeUses the networkConnects to a database
Not for
- Small projects that don't need complex deployment
- Users only interested in simple chat UI
Required permissions
- Network access for web search and connector sync
- Code execution in sandbox
- Storage and processing of user data
Risks and side effects
- Self-hosting requires operational expertise
- Potential data privacy and security concerns
- Costs if relying on external LLM providers
Setup
Before you start
Runtime：Docker · Python
Deploy with a single command: curl -fsSL https://onyx.app/install_onyx.sh | bash. See docs for detailed deployment guides.
Check that it works
This repository is a full AI platform rather than a single MCP server; verify by deploying via the README's one-line install script, opening the Onyx Chat UI, and confirming a configured LLM provider responds.
Troubleshooting
- Check resource requirements for minimum specs
- Inspect logs to diagnose deployment issues
- Ensure network connectivity
Things to try
Once connected, you can ask your AI assistant things like:
- Index our internal knowledge base via Onyx connectors and answer questions against it
- Build a custom agent in Onyx that uses web search and RAG to answer my questions
- Run a deep research report on a topic and generate a downloadable artifact
Use cases
Enterprise knowledge Q&A
Deep research
Custom AI agents
Integration with external applications
Overview
Onyx is the application layer for LLMs, offering a feature-rich interface that anyone can host. It enhances LLMs with advanced capabilities like RAG, web search, code execution, file creation, and deep research. Connect your applications with over 50+ indexing-based connectors out of the box or via MCP. Supports deployment via Docker, Kubernetes, Helm/Terraform, with standard and lite modes.
Similar servers
Hevy MCP
77 · B
Manage your Hevy workout data from AI assistants via MCP.
Firecrawl MCP Server
75 · B
Firecrawl's official MCP server for web search, scraping, and structured extraction for AI agents
BioMCP
75 · B
One binary. One grammar. Evidence from the biomedical sources you already trust.
Compartment
75 · B
Encrypted, fully offline long-term memory for AI agents
Source revision d6f2ee08f993 Data synced 2026-10-12 Read the FMRS scoring method