Tentra MCP

What rdanieli/tentra-mcp MCP server does

rdanieli/tentra-mcp MCP server connects an AI coding agent to two related capabilities: architecture modeling and a persistent code graph. An agent can turn a natural-language system description into a typed architecture diagram, update saved designs, inspect them, create flow visualizations, and apply nine listed quality checks, including orphan, single-point-of-failure, and god-service detection.

The architecture side also supports codebase analysis, diagram-to-code drift detection, and specification alignment checks. Designs can be exported to the formats listed by the project, including Java Spring Boot, Node.js Fastify, Python FastAPI, Go chi, Rust Axum, .NET ASP.NET, Kotlin Ktor, PHP Laravel, Ruby Rails, Elixir Phoenix, Docker Compose, Mermaid, ADR, and Terraform.

The code-graph side indexes files, symbols, imports, and call edges. Agents can search symbols, find references, prepare rename plans, inspect neighboring symbols, explain code paths, identify highly connected nodes, rank quality hotspots, compare snapshots, and retrieve repository-wide explanations. Additional tools store service mappings, domain membership, contracts, architecture decisions, and ownership data.

How it works

The local process runs through the tentra-mcp npm package and communicates with an MCP client over stdio. npx tentra-mcp init --hook can write configuration for supported installed clients, add a Git post-commit hook, and derive a repository identifier from the Git remote. The hook triggers background re-indexing after commits, keeping the graph updated without a separate manual indexing step.

For architecture operations and hosted access, the server uses the Tentra service at trytentra.com. The README describes both browser-based GitHub device-flow authentication and API-key authentication. Codebase scanning requires the local installation rather than the hosted SSE configuration. An agent invokes the available MCP tools; the server handles indexing jobs and persistence for the resulting graph data.

Setup and configuration

Install the local server with npx tentra-mcp. For repository-aware setup, run the project’s initialization command from the repository and optionally include --hook. The generated configuration can then be reloaded in the MCP client. The README names Cursor, Claude Code, Codex, and Windsurf as supported configuration targets.

The first local tool call can open GitHub device-flow authentication, with credentials saved under ~/.tentra/credentials. Alternatively, the command accepts an API key using --key. The SSE setup uses an endpoint at https://trytentra.com/api/mcp with the key supplied as a query parameter. The README’s examples use a placeholder key, so that value must be replaced before use.

Tools and capabilities

The listed tool groups cover:

• Architecture creation, updates, retrieval, listing, codebase analysis, linting, synchronization, alignment checks, exporting, and flow creation.
• Code-graph indexing, continuation of indexing jobs, semantic annotations, and job-status checks.
• Symbol search, reference lookup, rename planning, graph traversal, code explanations, similarity search, embeddings, snapshots, god-node detection, and hotspot analysis.
• Service-to-code mapping, domain assignment, contract recording and lookup, ADR recording and linking, and ownership resolution.

These capabilities let an agent move between a high-level system design and implementation-level evidence rather than relying only on repeated source-file searches.

Limitations and notes

rdanieli/tentra-mcp MCP server is not itself the hosted Tentra web application or API; the repository contains the open-source MCP server, while the API and web app are separate hosted services. The hosted SSE option does not provide the local codebase-scanning path described for the npm process.

The README lists 36 tools, while the supplied product description says 35, so the exact total should be verified against the current release. Authentication is required for the documented hosted or local service workflows, either through GitHub device flow or an API key. The project is licensed under MIT.