- CollectionSearch & Web Intelligence
- Classificationagent-native
- License⚠️ No repository license detected as of 2026-09-26 (GitHub metadata)
- MCPAvailable
- Agent SkillUnavailable
- InterfacesMCP · http · API
- Verified2026-09-26
What it does
contextX turns two hosted search modes into MCP tools. grok_search handles fast everyday retrieval, while grok_deep_search uses a multi-agent deep-search model for slower, broader investigation and cross-referencing. The server forwards MCP queries to OpenAI-compatible Responses API upstreams and continuously consumes their stream before returning the generated content as the MCP tool result, reducing long-search gateway timeout risk.
The public endpoint fixes the upstream models and requires no caller credential. Operators who need their own keys or endpoint policy can self-host the Rust server by supplying normal- and deep-search Responses API URLs and bearer tokens through environment variables.
Agent-facing interfaces
- Interaction pattern: remote MCP over Streamable HTTP
- Catalog protocols: MCP, http, API
- MCP: ✅ Available — public remote server
- Agent Skill: ⚠️ Not yet published — the repository provides MCP instructions, but no SKILL.md package.
- Onboarding type:
mcp
How agents onboard
Interaction pattern: remote MCP over Streamable HTTP
claude mcp add --transport http contextX https://mcp.twitter.monster/mcp
See the full onboarding section in the dossier.
Protocol surface
| Interface | Detail |
|---|---|
| Remote MCP | Streamable HTTP at https://mcp.twitter.monster/mcp |
| MCP tools | grok_search({query}), grok_deep_search({query}) |
| Health endpoint | /health on a self-hosted instance |
| Upstream protocol | OpenAI-compatible Responses API with streaming, configured independently for normal and deep search |
| Self-hosted runtime | Rust server started with cargo run; default MCP listener 127.0.0.1:3000/mcp |
Verification
Last verified 2026-09-26.
Sources:
- https://mcp.twitter.monster/mcp
- https://api.github.com/repos/akasakariko/ContextX