Tabby MCP Server
Official Tabby MCP server enabling AI agents to interact with Tabby — the self-hosted, open-source AI coding assistant — querying Tabby for code completions, accessing repository context indexed by Tabby, retrieving code search results, and integrating Tabby's local code intelligence into agent-driven development workflows without sending code to external services.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Self-hosted. Code stays on-premises. API token. HTTPS for remote. Official Tabby MCP — privacy-first.
⚡ Reliability
Best When
A team self-hosts Tabby for privacy-first AI code assistance and needs to integrate it with agent workflows — keeping code on-premises while leveraging AI coding capabilities.
Avoid When
Your team doesn't have a Tabby deployment or prefers cloud-based coding assistants.
Use Cases
- • Querying Tabby for code completions with full repo context from coding agents
- • Searching indexed codebase for relevant code patterns from code review agents
- • Accessing self-hosted AI code assistance from privacy-sensitive development agents
- • Integrating Tabby's code indexing with external agent workflows from developer agents
- • Building AI coding assistants for air-gapped environments from enterprise agents
- • Leveraging local code intelligence without cloud API costs from cost-conscious agents
Not For
- • Teams not using Tabby (use GitHub Copilot, Codeium, or other AI coding MCPs)
- • Cloud-only setups (Tabby requires self-hosted deployment)
- • Non-code use cases (Tabby is specialized for code intelligence)
Interface
Authentication
Tabby server API token required. Self-hosted Tabby deployment must be running and accessible. Configure TABBY_HOST and TABBY_AUTH_TOKEN.
Pricing
Tabby is free open source. Self-hosted requires GPU/CPU compute. MCP server is free. Tabby Cloud has paid managed plans.
Agent Metadata
Known Gotchas
- ⚠ Tabby server must be running and indexed — initial indexing of large repos takes significant time
- ⚠ GPU recommended for reasonable inference speed — CPU-only Tabby is slow
- ⚠ Tabby model selection affects code quality — choose model appropriate for target languages
- ⚠ Repository indexing via Tabby scheduler must complete before search is effective
- ⚠ Official Tabby MCP — maintained by TabbyML team and generally well-maintained
- ⚠ HTTPS recommended for production Tabby deployments — plain HTTP acceptable for local use
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Tabby MCP Server.
Scores are editorial opinions as of 2026-03-06.