nexus-agents
Intelligent orchestration platform that routes tasks to the best AI model (Claude, Codex, Gemini, OpenCode) using LinUCB bandits, validates through consensus voting, and learns from outcomes. 29 MCP tools, dev pipeline, 8 memory backends.
Nexus Agents
The intelligence layer between you and your AI coding tools
Why Nexus Agents?
nexus-agents makes your AI coding tools work together intelligently. It coordinates Claude, Codex, Gemini, and OpenCode — routing each task to the best model using data-driven algorithms, validating outputs through multi-model consensus voting, and continuously improving through outcome-driven learning. Connect it to any MCP-compatible editor (Claude Code, Cursor, VS Code) and it handles the rest.
What it does for you:
- Routes intelligently — LinUCB bandit + TOPSIS scoring + adaptive bonuses pick the right model for each task, learned from real outcomes
- Enforces quality — consensus voting (6 strategies including Bayesian higher-order), QA review loops, security scans with SARIF
- Learns over time — 5 memory backends (session, belief, agentic, adaptive, typed) track what works, feeding routing, planning, and research decisions
- Runs a full dev pipeline — research papers, plan architecture, vote on proposals, decompose into tasks, implement, QA review, ship
- Connects everything — 32 MCP tools, 9 research sources, graph workflows, checkpoint/resume, GitHub/GitLab issue tracking
You: "Review this code for security and performance"
↓
CompositeRouter selects best CLI per category → Security Expert + Code Expert
↓
Consensus-validated response — outcomes feed back into routing for next time
What it is NOT:
- Not an autonomous agent — humans stay in the loop via votes and harness mode
- Not a chat framework — it orchestrates real CLI tools with real file I/O
- Not a model API proxy — the intelligence IS the routing, quality gates, and learning
Architecture at a Glance
┌─────────────────────────────────┐
│ Your IDE / CLI │
│ (Claude Code, Cursor, VS Code) │
└──────────────┬──────────────────┘
│ MCP Protocol
┌──────────────▼──────────────────┐
│ nexus-agents server │
│ │
│ ┌──────────┐ ┌──────────────┐ │
│ │ 32 MCP │ │ Dev Pipeline │ │
│ │ Tools │ │ research→plan │ │
│ └────┬─────┘ │ →vote→impl │ │
│ │ │ →QA→ship │ │
│ ┌────▼─────┐ └──────────────┘ │
│ │Composite │ │
│ │Router │ ┌──────────────┐ │
│ │(9 stages)│ │ 5 Memory │ │
│ └────┬─────┘ │ Backends │ │
│ │ └──────────────┘ │
└───────┼─────────────────────────┘
┌────────────┼────────────┐
▼ ▼ ▼
┌────────┐ ┌────────┐ ┌──────────┐
│ Claude │ │ Gemini │ │ Codex │ ...
│ CLI │ │ CLI │ │ CLI │
└────────┘ └────────┘ └──────────┘
Quick Start (2 minutes)
1. Install
npm install -g nexus-agents
Or as a Claude Code plugin (single-command install from the official marketplace):
/plugin install nexus-agents
See docs/getting-started/PLUGIN_INSTALL.md for plugin-specific setup, or llms-install.md for the short install guide an AI agent can follow.
2. Verify
nexus-agents doctor
3. Use
With Claude Code (recommended):
nexus-agents setup # Auto-configures MCP server
Then in Claude: "orchestrate: Review this PR for issues"
Standalone CLI:
export ANTHROPIC_API_KEY=your-key
nexus-agents orchestrate "Explain the architecture of this codebase"
Security: In default MCP mode, the server communicates only via stdio with the parent process (no network exposure). The REST API (opt-in) auto-generates an API key on first start. For network-exposed deployments, set
NEXUS_AUTH_ENABLED=true. See SECURITY.md.
Capabilities
| Category | Details |
|---|---|
| Intelligent Routing | 9-stage CompositeRouter: budget-aware, LinUCB bandit, TOPSIS multi-criteria, preference-trained, weather-adaptive. Learns from outcomes. |
| Multi-Expert Orchestration | 11 built-in expert types (code, architecture, security, testing, docs, devops, research, PM, UX, infrastructure, data-visualization) coordinated by TechLead/Orchestrator agents |
| Consensus Voting | 6 strategies: simple_majority, supermajority, unanimous, higher_order (Bayesian correlation-aware), opinion_wise, proof_of_learning |
| Development Pipeline | Research → Plan → Vote → Decompose → Implement → QA → Security. Three modes: autonomous, harness (caller implements), dry-run |
| Memory & Learning | 5 user-facing backends (session, belief, agentic, adaptive, typed). Cross-session persistence. Outcomes feed routing. |
| Research System | 9 discovery sources (arXiv, GitHub, Semantic Scholar, etc). Auto-catalog, quality scoring, synthesis into topic clusters |
| Security | Sandboxing (Docker/policy), trust classification, SARIF parsing, input sanitization, red team pipeline, firewall |
| Graph Workflows | DAG-based workflow execution with checkpoint/resume, state reduction, and event hooks |
| 32 MCP Tools | Agent management, workflow execution, research, memory, codebase intelligence, repo analysis, consensus, operations |
Available Experts
| Expert | Specialization |
|---|---|
| Code | Implementation, debugging, optimization |
| Architecture | System design, patterns, scalability |
| Security | Vulnerability analysis, secure coding |
| Testing | Test strategies, coverage, test generation |
| Documentation | Technical writing, API docs |
| DevOps | CI/CD, deployment, infrastructure |
| Research | Literature review, state-of-the-art analysis |
| PM | Product management, requirements, priorities |
| UX | User experience, usability, accessibility |
| Infrastructure | Server management, bare metal, networking |
| Data Viz | Charts, dashboards, visual data presentation |
Supported CLIs & Providers
Nexus-agents routes tasks through 5 CLI adapters, each connecting to major AI providers:
| CLI | Provider | Best For |
|---|---|---|
| claude | Anthropic (Claude) | Complex reasoning, analysis |
| gemini | Google (Gemini) | Long context, multimodal |
| codex | OpenAI (Codex CLI) | Code generation, reasoning |
| codex-mcp | OpenAI (Codex MCP) | MCP-native Codex integration |
| opencode | Custom OpenAI-compat | Custom endpoints, local models |
CLI Commands
nexus-agents # Start MCP server (default)
nexus-agents doctor # Check installation health
nexus-agents setup # Configure Claude CLI integration
nexus-agents orchestrate "..." # Run task with experts
nexus-agents vote "proposal" # Multi-agent consensus voting
nexus-agents review <pr-url> # Review a GitHub PR
nexus-agents expert list # List available experts
nexus-agents workflow list # List workflow templates
nexus-agents config init # Generate config file
nexus-agents fitness-audit # Run fitness score audit
nexus-agents research query # Query research registry
nexus-agents --help # Full command list
See docs/ENTRYPOINTS.md for the complete CLI reference (28+ commands).
MCP Tools
When running as an MCP server, the following tools are available:
| Tool | Description |
|---|---|
orchestrate | Task orchestration with Orchestrator coordination |
create_expert | Create a specialized expert agent |
execute_expert | Execute a task using a created expert |
run_workflow | Execute a workflow template |
delegate_to_model | Route task to optimal model |
consensus_vote | Multi-model consensus voting on proposals |
list_experts | List available expert types |
list_workflows | List available workflow templates |
research_query | Query research registry (status, overlap, stats, search) |
research_add | Add paper to registry by arXiv ID |
research_discover | Discover papers/repos from external sources |
research_analyze | Analyze registry for gaps, trends, coverage |
research_catalog_review | Review auto-cataloged research references |
memory_query | Query across all memory backends |
memory_stats | Memory system statistics dashboard |
memory_write | Write to typed memory backends |
weather_report | Multi-CLI performance weather report |
issue_triage | Triage GitHub issues with trust classification |
run_graph_workflow | Execute graph-based workflows with checkpointing |
execute_spec | Execute AI software factory spec pipeline |
registry_import | Generate draft model registry entry |
query_trace | Query execution traces for observability |
repo_analyze | Analyze GitHub repository structure |
repo_security_plan | Generate security scanning pipeline for a repo |
research_add_source | Add non-paper source (GitHub repo, tool, blog) |
research_synthesize | Synthesize registry into topic clusters with themes |
extract_symbols | Extract code symbols from source files for analysis |
search_codebase | Search codebase for patterns, symbols, or text |
run_dev_pipeline | Full dev pipeline: research, plan, vote, implement, QA |
run_pipeline | Execute a pipeline plugin by name with typed input |
Configuration
Environment Variables:
| Variable | Description |
|---|---|
ANTHROPIC_API_KEY | Claude API key |
OPENAI_API_KEY | OpenAI API key |
GOOGLE_AI_API_KEY | Gemini API key |
NEXUS_LOG_LEVEL | Log level (debug/info/warn/error) |
Generate config file:
nexus-agents config init # Creates nexus-agents.yaml
Documentation
| Topic | Link |
|---|---|
| Full CLI Reference | docs/ENTRYPOINTS.md |
| Architecture | docs/architecture/README.md |
| Contributing | CONTRIBUTING.md |
| Coding Standards | CODING_STANDARDS.md |
| Quick Start Guide | QUICK_START.md |
Development
git clone https://github.com/williamzujkowski/nexus-agents.git
cd nexus-agents
pnpm install
pnpm build
pnpm test
Requirements: Node.js 22.x LTS, pnpm 9.x
Contributing
- Fork the repository
- Create a feature branch (
git checkout -b feat/amazing-feature) - Commit with conventional commits (
feat(scope): add feature) - Open a Pull Request
See CONTRIBUTING.md for details.
License
MIT - See LICENSE
Built with Claude Code
相关服务器
Alpha Vantage MCP Server
赞助Access financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
Dev Manager
A development management tool for project planning, task management, and development workflows.
Cycode
Boost security in your dev lifecycle via SAST, SCA, Secrets & IaC scanning with Cycode.
Talk to Figma MCP
A server for integrating with Figma, allowing you to interact with your design files.
mcp-diagnostics
All-in-one website diagnostics: DNS, SSL certs, HTTP headers, security audit (A-F grade), WHOIS, tech stack detection. No API keys needed.
MCP Performance Analysis Server
A server for detecting critical performance issues in code, providing concise analysis and output.
Figma MCP Server
Enables AI assistants to interact with Figma via WebSocket for reading data and design analysis.
MediaWiki
Interact with MediaWiki installations through the MediaWiki API as a bot user.
EdgeOne Pages MCP
An MCP server implementation using EdgeOne Pages Functions for intelligent chat applications.
SDD MCP
Provides Seam-Driven Development tools for AI-assisted software development.
BCMS MCP
Give me a one - two sentence description of the BCMS MCP # MCP The BCMS Model Context Protocol (MCP) integration enables AI assistants like Claude, Cursor, and other MCP-compatible tools to interact directly with your BCMS content. This allows you to create, read, and update content entries, manage media files, and explore your content structure—all through natural language conversations with AI. ## What is MCP? The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) is an open standard developed by Anthropic that allows AI applications to securely connect to external data sources and tools. With BCMS MCP support, you can leverage AI assistants to: - Query and explore your content structure - Create new content entries with AI-generated content - Update existing entries - Manage your media library - Get intelligent suggestions based on your content model --- ## Getting Started ### Prerequisites 1. A BCMS account with an active instance 2. An MCP key with appropriate permissions 3. An MCP-compatible client (Claude Desktop, Cursor, or any MCP client) ### Step 1: Create an MCP Key 1. Navigate to your BCMS dashboard 2. Go to Settings → MCP 3. Click Create MCP Key 4. Configure the permissions for templates you want the AI to access:GET: Read entries 5. POST: Create entries 6. PUT: Update entries 7. DELETE: Delete entries Note: Right now, MCP only supports creating, reading and updating content. ### Step 2: Configure Your MCP Client You can find full instructions for integrating BCMS with your AI tools right inside BCMS, on the MCP page. But in general, installing BCMS MCP works in a standard way: ``` { "mcpServers": { "bcms": { "url": "https://app.thebcms.com/api/v3/mcp?mcpKey=YOUR_MCP_KEY" } } } ``` ## Available Tools Once connected, your AI assistant will have access to the following tools based on your MCP key permissions: ### Content Discovery #### list_templates_and_entries Lists all templates and their entries that you have access to. This is typically the first tool to call when exploring your BCMS content. Returns: - Template IDs, names, and slugs - Entry IDs with titles and slugs for each language Example prompt: "Show me all the templates and entries in my BCMS" --- ### Entry Management #### list_entries_for_{templateId} Retrieves all entries for a specific template with full content data. A separate tool is generated for each template you have access to. Returns: - Complete entry data including all meta fields - Content in all configured languages - Entry statuses Example prompt: "List all blog posts from my Blog template" --- #### create_entry_for_{templateId} Creates a new entry for a specific template. The input schema is dynamically generated based on your template's field structure. Input: - statuses: Array of status assignments per language - meta: Array of metadata for each language (title, slug, custom fields) - content: Array of content nodes for each language Example prompt: "Create a new blog post titled 'Getting Started with BCMS' with a brief introduction paragraph" --- #### update_entry_for_{templateId} Updates an existing entry for a specific language. Input: - entryId: The ID of the entry to update - lng: Language code (e.g., "en") - status: Optional status ID - meta: Updated metadata - content: Updated content nodes Example prompt: "Update the introduction paragraph of my 'Getting Started' blog post" --- ### Media Management #### list_all_media Lists all media files in your media library. Returns: - Media IDs, names, and types - File metadata (size, dimensions for images) - Parent directory information Example prompt: "Show me all images in my media library" --- #### list_media_dirs Lists the directory structure of your media library. Returns: - Hierarchical directory structure - Directory IDs and names Example prompt: "Show me the folder structure of my media library" --- #### create-media-directory Creates a new directory in your media library. Input: - name: Name of the directory - parentId: Optional parent directory ID (root if not specified) Example prompt: "Create a new folder called 'Blog Images' in my media library" --- #### request-upload-media-url Returns a URL you use to upload a file (for example via POST with multipart form data), which avoids pushing large binaries through the MCP tool payload. You still need a valid file name and MIME type when uploading, as described in the tool response. Availability: Only when the MCP key has Can mutate media enabled. Example prompt: “Give me an upload URL for a new hero image, then tell me how to upload it.” Input: - fileName: Name of the file with extension - fileData: Base64-encoded file data (with data URI prefix) - parentId: Optional parent directory ID Example prompt: "Upload this image to my Blog Images folder" --- ### Linking Tools #### get_entry_pointer_link Generates an internal BCMS link to an entry for use in content. Input: - entryId: The ID of the entry to link to Returns: - Internal link format: entry:{entryId}@*_{templateId}:entry Example prompt: "Get me the internal link for the 'About Us' page entry" --- #### get_media_pointer_link Generates an internal BCMS link to a media item for use in content. Input: - mediaId: The ID of the media item Returns: - Internal link format: media:{mediaId}@*_@*_:entry Example prompt: "Get the link for the hero image so I can use it in my blog post" --- ## Content Structure ### Entry Content Nodes When creating or updating entries, content is structured as an array of nodes. Supported node types include: Type Description paragraph Standard text paragraph heading Heading (h1-h6) bulletList Unordered list orderedList Numbered list listItem List item codeBlock Code block with syntax highlighting blockquote Quote block image Image node widget Custom widget with props ### Example Content Structure ``` { "content": [ { "lng": "en", "nodes": [ { "type": "heading", "attrs": { "level": 1 }, "content": [ { "type": "text", "text": "Welcome to BCMS" } ] }, { "type": "paragraph", "content": [ { "type": "text", "text": "This is your first paragraph." } ] } ] } ] } ``` ## Security & Permissions ### MCP Key Scopes Your MCP key controls what the AI can access: - Template Access: Only templates explicitly granted in the MCP key are visible - Operation Permissions: Each template can have independent GET/POST/PUT/DELETE permissions - Media Access: Media operations are controlled separately ### Best Practices 1. Principle of Least Privilege: Only grant the permissions needed for your use case 2. Separate Keys: Create different MCP keys for different purposes or team members 3. Regular Rotation: Periodically rotate your MCP keys ## Use Cases ### Content Creation Workflows Blog Post Creation "Create a new blog post about the benefits of headless CMS. Include an introduction, three main benefits with explanations, and a conclusion. Use the Blog template." Product Updates "Update the price field for all products in the Electronics category to apply a 10% discount" ### Content Exploration Content Audit "List all blog posts that don't have a featured image set" Translation Status "Show me which entries are missing German translations" ### Media Organization Library Cleanup "Show me all unused images in the media library" Folder Setup "Create folder structure for: Products > Categories > Electronics, Clothing, Home" ## Troubleshooting ### Common Issues #### "MCP key not found" - Verify your MCP key format: keyId.keySecret.instanceId - Ensure the MCP key hasn't been deleted or deactivated - Check that you're using the correct instance #### "MCP key does not have access to template" - Review your MCP key permissions in the dashboard - Ensure the required operation (GET/POST/PUT/DELETE) is enabled for the template #### Session Expired - MCP sessions may timeout after periods of inactivity - Simply start a new conversation to establish a fresh session ### Getting Help - Documentation: [thebcms.com/docs](https://thebcms.com/docs) - Support: [[email protected]](mailto:[email protected]) - Community: [Join BCMS Discord](https://discord.com/invite/SYBY89ccaR) for community support ## Technical Reference ### Endpoint POST https://app.thebcms.com/api/v3/mcp?mcpKey={MCP_KEY} ### Transport BCMS MCP uses the Streamable HTTP transport with session management. Sessions are maintained via the mcp-session-id header. ### Response Format All tools return structured JSON responses conforming to the MCP specification with: - content: Array of content blocks - structuredContent: Typed response data ## Rate Limits MCP requests are subject to the same rate limits as API requests: - Requests are tracked per MCP key - Contact support if you need higher limits for production workloads