Jarvis Registry
A unified MCP gateway for governed enterprise-tool access by copilots and autonomous agents.
- Source repo
- ascending-llc/jarvis-registry
- Stars
- ★ 3.8k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 40/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A platform team wants Cursor, Claude Desktop, GitHub Copilot, or VS Code to reach internal tools through one authenticated MCP entry point.
- Not a fit if
- Teams that need fully offline operation without identity-provider credentials
- Teams wanting a managed cloud service instead of self-hosting
- Users who need a Chinese-language UI or documentation
- Source review
- 40/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Jarvis Registry is an open-source enterprise MCP (Model Context Protocol) and A2A agent gateway and workflow orchestration platform from ASCENDING Inc. Its MCP Gateway & Reverse Proxy provides a single authenticated entry point for MCP clients and agents over SSE or Streamable HTTP. The platform can register and manage autonomous agents, with orchestrator agents coordinating worker agents through that same gateway. Its identity layer supports OAuth 2.0/OIDC integration with Keycloak, Amazon Cognito, and Microsoft Entra ID, while its ACL engine enforces scope-, role-, and individual-tool-level permissions. A full local stack can be deployed with Docker Compose, and the project also documents deployment guides for AWS EKS, Azure AKS, and GCP GKE.
After deployment, the MCP Gateway & Reverse Proxy accepts AI-client and agent traffic using MCP over SSE or Streamable HTTP and acts as the unified authentication entry point. It integrates OAuth 2.0/OIDC with Keycloak, Amazon Cognito, or Microsoft Entra ID, then uses its ACL engine to enforce access by scope, role, and individual tool. The registry registers and manages autonomous A2A agents; orchestrator agents can coordinate worker agents through the same secure gateway. It also performs semantic search across skills, descriptions, and tags to discover an MCP server or A2A agent, and provides request logging, OpenTelemetry tracing, and Prometheus metrics.
- A platform team wants Cursor, Claude Desktop, GitHub Copilot, or VS Code to reach internal tools through one authenticated MCP entry point.
- A security team needs AI-tool permissions tied to Keycloak, Amazon Cognito, or Microsoft Entra ID and enforced down to individual tools.
- An automation team is building an autonomous-agent workflow that needs registered A2A agents and an orchestrator coordinating worker agents.
- An organization with multiple MCP servers needs runtime discovery based on skills, descriptions, and tags.
- An operations team needs request logs, OpenTelemetry traces, and Prometheus metrics for copilot and agent traffic.
How do you install or deploy this agent?
Prerequisites explicitly shown are Python, uv, Docker Compose, and identity-provider credentials. Run:
git clone https://github.com/ascending-llc/jarvis-registry.git
cd jarvis-registry
cp .env.example .env
# Configure identity-provider credentials in .env
uv sync --all-packages
source .venv/bin/activate
docker compose --profile full up -dWhen the full local stack is running, open http://localhost:80.
How do you use this agent?
The first documented working action is to configure identity-provider credentials in .env, start the full Docker Compose profile, and open http://localhost:80. The README states that the gateway accepts MCP over SSE or Streamable HTTP and names Cursor, Claude Desktop, GitHub Copilot, and VS Code as integrations. It does not provide copyable MCP client configuration, agent-registration steps, tool-call examples, or API/CLI invocation details, so those first integration actions cannot be reproduced from the supplied material alone.
What are this agent's strengths and limitations?
- One MCP Gateway & Reverse Proxy covers both SSE and Streamable HTTP instead of requiring separate client-facing entry points.
- OAuth 2.0/OIDC integrations are explicitly named for Keycloak, Amazon Cognito, and Microsoft Entra ID, with ACL enforcement by scope, role, and individual tool.
- The same platform covers MCP-server discovery plus A2A agent registration, management, and orchestration.
- Request logging, OpenTelemetry tracing, and Prometheus metrics are built into the stated observability model.
- A full local deployment depends on Python, uv, Docker Compose, and configured identity-provider credentials.
- The README does not provide concrete MCP client configuration, API/CLI commands, agent-registration procedures, or end-to-end tool-call examples.
- Only Keycloak, Amazon Cognito, and Microsoft Entra ID are explicitly named as identity integrations; support for other providers is not evidenced.
- Although EKS, AKS, and GKE deployment guides are listed, the supplied material does not specify their configuration, scaling model, or failure-handling behavior.
How does this agent compare with similar options?
The README says Jarvis Registry builds on foundational work from agentic-community/mcp-gateway-registry and positions this project as an enterprise evolution with identity, access control, observability, and A2A orchestration capabilities.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Jarvis Registry This agent | 40 · Major gaps | Self-hosted serviceFree | ★ 3.8k | today | Python | — |
| AURA SRE Agent Platform | 71 · Some gaps | CLIFree + model costs | ★ 410 | 6d ago | Rust | OpenAI API |
| MCPJungle | 65 · Some gaps | Self-hosted serviceFree | ★ 1.6k | 2d ago | Go | Codex · Claude.ai |
| ContextForge Gateway | 49 · Major gaps | Self-hosted serviceFree | ★ 4.8k | today | Python | Claude.ai · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the project claims to provide OAuth/OIDC, ACL engine, and audit logging, but no implementation details or security audit reports are provided. Dependency security: pyproject.toml lists dependencies like cryptography, but no vulnerability scanning or lock files are provided. External effects: as a gateway, it may affect external systems, but this is not explicitly stated. Rollback: not mentioned. Source attribution: README acknowledges based on agentic-community/mcp-gateway-registry, but no specific contributor information is provided. Therefore, most criteria score 1 due to insufficient evidence or mere assertion.
The project structure is clear, with multiple subprojects (registry, auth-server, registry-pkgs), and pyproject.toml defines test tasks, indicating some consistency. However, dependency availability: no lock files or dependency manifests are provided, so it is unclear if dependencies are obtainable. Failure messages: no examples of error handling or user prompts are provided. Thus, self-consistency scores 2, the other two score 1.
README clearly targets enterprise AI copilots and autonomous agents, and provides multiple deployment scenarios (AWS, Azure, GCP, Docker Compose), indicating some adaptability. However, capability boundaries are not clearly defined, and trigger precision: no specific trigger conditions or use cases are provided. Environment fit: multiple cloud deployment guides are provided, but detailed configuration requirements are not. Therefore, audience and scenarios score 2, the other two score 1.
Information architecture: README provides a clear documentation structure, including quick start, documentation links, etc. Install notes: basic quick start steps are provided, but detailed configuration guides are not. Naming stability: project name and version are defined in pyproject.toml, but no naming conventions are provided. Examples and FAQ: README provides a demo video link, but no FAQ. Known limitations: not mentioned. License: Apache-2.0, compliant. Versioning and changelog: version number in pyproject.toml, but no changelog. Maintenance responsibility: README provides contribution guide and security report email, but maintainers are not clearly identified. Therefore, most criteria score 1 or 2.
Output usability: project provides UI and API, but no specific output examples. Marginal value: project claims to solve enterprise AI integration problems, but no comparison with other solutions. Cost-benefit: no performance or resource consumption data. Therefore, marginal value scores 2, the other two score 1.
Claim traceability: feature claims in README are not supported by specific implementation or test evidence. Cross-source corroboration: project claims to be based on another project, but no verification is provided. Fact-inference separation: descriptions in README are mostly assertions, not distinguishing facts from inferences. Therefore, all criteria score 1.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- The project claims enterprise-grade security features, but no implementation details or security audits are provided; evaluate with caution.
- Dependency security lacks lock files or vulnerability scanning, posing supply chain risks.
- No rollback mechanism is provided, which may affect disaster recovery.
- Known limitations are not mentioned, potentially hiding underlying issues.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
asc0.5.17 Jarvis Registry asc0.5.17
Release notes excerpt
🐛 Bug Fixes - Fixes Google Workspace group access grants not reaching users first created by Chat; Google users are now matched to groups by email, while Entra users still match by their directory object ID. (#654) - Preserves existing embeddings when updating server and agent metadata such as enabled state, tags and runtime version, which previously failed with a 500 error in environments without static AWS keys or silently overwrote the stored vector. (#653) …View on GitHub ↗ -
asc0.5.16 Jarvis Registry asc0.5.16
Release notes excerpt
✨ Features - Adds a migration so API key fields saved on MCP servers are stored in a consistent format, returned in the expected format on read, and default correctly when switching a server from OAuth to API Key authentication. (#643) - Improves discovery guidance in the Registry MCP gateway: empty discovery results now return hints tailored to the search that ran, and the MCP instructions and client documentation are simplified. (#638) 🐛 Bug Fixes …View on GitHub ↗ -
asc0.5.15 Jarvis Registry asc0.5.15
Release notes excerpt
🔧 Refactoring & Performance - Moves embedding reindex and skill sync onto a shared lease mechanism so only one instance runs each job at a time, and fixes a case where a failed source write after job completion left a skill source stuck in a syncing state. (#617) - Pins pre-commit Python hook environments to Python 3.12 so hook installation works with newer system Pythons. (#631) - Resolves intermittent failure in the ACL role reload throttle test by making its clock deterministic. (#641) …View on GitHub ↗