Orchestrator MCP Server
# Orchestrator MCP Server
Model Context Protocol server that bridges Claude Code to the Cloud Orchestrator API.
## Installation
Already installed and configured in Claude Code MCP settings.
## Tools Available
| Tool | Description |
|------|-------------|
| think | Deep reasoning via cognitive orchestrator |
| consensus | Multi-AI consensus (6 providers) |
| fast_ai | Ultra-fast response via Groq |
| web_search | Real-time web search via Perplexity |
| scrape_url | Web scraping via Firecrawl |
| run_python | Execute Python in E2B sandbox |
| run_javascript | Execute JavaScript in E2B sandbox |
| mem0_add | Store in long-term memory |
| mem0_search | Search long-term memory |
| neo4j_query | Query knowledge graph |
| railway_status | Check deployment status |
| railway_set_variable | Set env var on Railway |
| railway_redeploy | Trigger deployment |
| github_repos | List GitHub repos |
| github_file | Read file from repo |
| health | Check all service statuses |
| wolfram_query | Query Wolfram Alpha |
| wolfram_calculate | Math calculation |
| wolfram_convert | Unit conversion |
| openrouter_chat | Query any AI model |
| openrouter_models | List available models |
| sentry_stats | Error tracking stats |
| sentry_errors | Recent errors |
| api_keys_summary | API keys status |
| api_keys_inject | Inject API key to Railway |
## Configuration
The MCP server connects to:
- **Cloud Orchestrator:** https://web-production-bdfb4.up.railway.app
Environment variable (optional):
- `ORCHESTRATOR_URL` - Override default orchestrator URL
## Endpoint Mapping
| MCP Tool | API Endpoint |
|----------|-------------|
| run_python | POST /e2b/python |
| run_javascript | POST /e2b/javascript |
| mem0_add | POST /mem0/add |
| mem0_search | POST /mem0/search |
| think | POST /think |
| consensus | POST /ai/consensus |
## Recent Fixes (Jan 6, 2026)
- Fixed mem0_add endpoint: /mem0/memory -> /mem0/add
- Fixed mem0_search: Changed from GET with query params to POST
## Development
```bash
cd C:\Users\matt\orchestrator-mcp
npm install
node index.js # Runs as stdio MCP server
```
TDQS
Scored across 28 tools
The tool set has clear clusters for distinct domains like API key management, AI services, and Railway operations, but there is significant overlap within some clusters. For example, wolfram_calculate, wolfram_convert, and wolfram_query have overlapping purposes that could cause confusion, and fast_ai, openrouter_chat, and think all involve AI interactions with unclear boundaries. Descriptions help differentiate some tools, but misselection is possible due to this overlap.
Most tools follow a consistent snake_case naming pattern with clear verb_noun structures, such as api_keys_inject, railway_redeploy, and mem0_search. However, there are minor deviations like consensus, health, and think, which use single words without a noun, slightly breaking the pattern. Overall, the naming is mostly predictable and readable across the set.
With 28 tools, the count is borderline high for a single server, as it covers multiple broad domains like cloud orchestration, AI services, and data processing. This could feel heavy and overwhelming for agents, but it is reasonable given the server's apparent scope as an 'Orchestrator' integrating various services. A more focused scoping might reduce the count for better usability.
The tool surface covers a wide range of operations across its domains, such as CRUD-like actions for API keys and Railway services, and diverse AI and data processing capabilities. Minor gaps exist, like no explicit update or delete tools for some resources (e.g., GitHub files or Mem0 entries), but agents can likely work around these with available tools. The set supports core workflows without major dead ends.