Gephi MCP

Built for researchers working across network science and AI.

Status: public beta.APIs may change between minor versions.

Your AI assistant drives Gephi.Say what you want in plain language and the assistant builds, analyzes, styles, and exports publication-ready network maps.

It's a conversation, not a command line.The assistant explains what it's doing, checks its own maps before showing them, and teaches you to read what you're seeing. You can point back: select nodes in the Gephi window and ask "what did I select?"

Any data, any MCP client.Network files import directly; spreadsheets and other data become networks conversationally. Works with Claude Code, Claude Desktop, or any MCP-compatible assistant.

- 105 tools covering the whole workflow: build, analyze, style, lay out, filter, and export
- One-level undo: destructive operations snapshot the workspace automatically, sogephi_undobrings the graph back
- Layout quality is measured, not eyeballed: the graph profile flags heavy-tailed weights and hub-and-spoke wiring before a layout runs, visual QA scores how well communities separated and frames exports around the main cloud (ignoring runaway outlier nodes), and numerically exploded layouts are caught automatically instead of exported
- Interactive network view inside the chat (pan, zoom, hover, click a node to ask about it)
- Slash commands for common jobs:/analyze-network,/community-detection,/centrality,/visualize,/import-and-explore,/beautify,/verify-claim,/text-network,/teach,/counterfactual
- Specialized agents that run multi-step work in their own context and hand back just the result: independent claim verification, layout iteration, structural analysis, and text-network construction
- Two extra layouts beyond Gephi's own: by role played in the network, and by community (best for reply and retweet networks)
- Reads your selection in the Gephi window, so "what did I select?" just works
- Drives any layout or metric plugin installed from the Gephi plugin portal
- Imports GEXF, GraphML, GML, CSV, DOT, and Pajek; turns spreadsheets and other files into networks conversationally

Three components connect your AI assistant to Gephi Desktop:

Claude / AI Assistant │ MCP Protocol (stdio) │ MCP Server (Python) ← Translates MCP tool calls to HTTP │ HTTP API (localhost:8080) │ Gephi Plugin (Java) ← Runs inside Gephi Desktop │ Gephi Desktop ← Must be running first

Install the Gephi plugin plus your AI client's connection — the Claude Code plugin bundles the MCP server, so most users install just two things. Gephi Desktop must be running before using any tools.

Security note:the plugin's HTTP API binds to127.0.0.1only, validates the requestHostheader (DNS-rebinding defense), and sends no CORS headers — it is reachable only by local processes such as the MCP server, never by a web page. It is not authenticated, so do not expose port 8080 beyond localhost.

macOS note:older plugin versions could wedge Gephi during sustained writes against a large rendered graph (calls hang; onlygephi_health_checkanswers). Plugin 1.2.0 fixes the two causes on our side: writes now pause the renderer via Gephi's own viz-engine API, and a read-lock leak in the query endpoints (the main culprit) is closed. All lock waits are bounded, so a genuinely wedged Gephi returns an immediate "fully quit and reopen Gephi" error instead of hanging, andgephi_health_checkexposes lock probes (graph_lock,graph_lock_stats) that detect the condition. If you ever see persistent "Graph is busy" errors, restart Gephi — and make sure you are on plugin 1.2.2 or newer.

Verify:Open a browser tohttp://127.0.0.1:8080/health— you should see{"success": true}.

Use ONE connection method per app. If you previously addedgephi-mcptoclaude_desktop_config.jsonby hand, remove that entry before installing the bundle — otherwise Claude Desktop runs two copies of the server and every tool appears twice.

Add to your MCP configuration (claude_desktop_config.json):

Avoid installing with pip inside a project virtual environment. Inside an activated.venvthegephi-mcpcommand only exists on that venv'sPATH, and MCP clients launch servers outside your shell — the server fails with "Executable not found in $PATH" even thoughwhich gephi-mcpsucceeds. Useuvx/pipx, or point your MCP config at the venv executable's absolute path.

First confirm the MCP server actually connected. In Claude Code, run/mcp—gephi-mcpshould be listed asconnected. (In Claude Desktop, check that the tools appear in the tools menu.) If it shows a failure like "Executable not found in $PATH", the launcher (uv/uvxorgephi-mcp) isn't on the globalPATH— see the notes in Step 2.

Then, with Gephi running, ask your assistant:

It should callgephi_health_checkand confirm the connection. In Claude Code, try:

/gephi-network-analysis:import-and-explore path/to/your/graph.gexf

gephi-ai improves often, so keep it current. Most fixes and new tools arrive through updates. The health check tells you once per session when your install is behind the latest release and shows the matching update step, but that notice only appears on recent versions, so if you installed a while ago (or an older command like/teachis missing), run the update below once to catch up. Updating is safe to do anytime.

- Claude Code:claude plugin update gephi-network-analysis@gephi-ai, then start a new session.
- Claude Desktop (one-click bundle):download the newest.mcpbfrom Releases and double-click it again.
- Claude Desktop (config file):nothing to do — theuvxentry fetches the latest release each time you fully quit and reopen Claude Desktop.
- Switching connection methods:remove the old one first (bundle: uninstall in Settings > Extensions; config file: delete thegephi-mcpblock). Two methods at once means two servers and duplicated tools.
- Cowork:Cowork keeps its own copy of plugins, separate from Claude Code — updating one does not update the other. If Cowork's commands look older than this README (for example, no/teach), ask Cowork itself to update the gephi-network-analysis plugin, then fully quit and reopen the app.
- Gephi plugin:install the newest.nbmfrom Releases via Tools > Plugins > Downloaded and restart Gephi.

gephi_health_checkreports both the Gephi plugin version and the MCP server version, and says whether they are up to date, so you can see at a glance what you are running.

The plugin (claude-plugin/) goes beyond raw MCP tools:

1. gephi_create_project 2. gephi_import_file (your GEXF, GraphML, or CSV) 3. gephi_compute_degree 4. gephi_compute_modularity (resolution: 1.0) 5. gephi_color_by_partition (column: modularity_class) 6. gephi_size_by_ranking (column: degree) 7. gephi_run_layout (ForceAtlas 2, 1000 iterations) 8. gephi_export_png (3840x2160 for publication)

1. Import or build graph 2. gephi_compute_betweenness 3. gephi_compute_pagerank 4. gephi_run_layout (ForceAtlas 2) 5. gephi_color_by_ranking (column: betweenesscentrality) 6. gephi_size_by_ranking (column: pageranks) 7. gephi_query_nodes (find top-ranked nodes)

Reference guides are inclaude-plugin/skills/gephi/:

- SKILL.md— Workflow patterns, best practices, and critical gotchas
- references/tool-reference.md— Complete API reference for all 105 tools
- references/layout-guide.md— Layout algorithm selection and parameter tuning
- references/statistics-guide.md— Statistics interpretation guide

- Gephi Plugin: Java 11, NetBeans Platform, NanoHTTPD, Gson
- MCP Server: Python 3.10+, MCP SDK (FastMCP), httpx, Pydantic, defusedxml; vendored sigma.js + graphology for the in-chat viewer
- Target: Gephi 0.11.1, NetBeans RELEASE290

Building the Gephi plugin from source requires JDK 11+ and Maven:

cd gephi-mcp-plugin mvn clean package # output: target/gephi-mcp-<version>.nbm, auto-deployed to Gephi's modules dir

Fully quit and reopen Gephi after every rebuild, even mid-session.Gephi lazily loads plugin classes on first use; overwriting the jar file on disk while Gephi is still running an earlier build corrupts the classloader's view of any class not yet loaded, and the resulting crash (a raw dropped connection, no error message) can be hard to trace back to the real cause.

The MCP server is a standard Python package undermcp-server/(pytestfor tests,rufffor linting).

If you use or adapt this project in your work, please credit:

If you use this toolkit in your academic research, please cite:

The 1Password MCP server creates a bridge that allows MCP clients such as Codex and Kiro to manage your 1Password Environments with secure authorization prompts.

This is the 1st, easiest, and cheapest PPT, slides, presentation AI generation MCP Server in the world.

Persistent memory for any AI assistant. Zero token cost until recall. Stores memories in local SQLite, ranks by 6-factor scoring, returns results 79% smaller than JSON. Works with Claude, ChatGPT, Grok, Cursor, Windsurf, and any MCP client.

A MCP server that enables AI assistants to interact with Anki, the spaced repetition flashcard application.

Enables LLM clients to interact with macOS applications through AppleScript. Built using the @beyondbetter/bb-mcp-server library, this server provides safe, controlled execution of predefined scripts with optional support for arbitrary script execution.

An MCP server for WordPress plugin audits

Turn your AI assistant into a digital marketing hub that creates, organizes, and analyzes links and QR Codes on demand.

Connect AI clients to Cal.com scheduling through the Model Context Protocol using the hosted server at mcp.cal.com or a local instance.