Skip to content
140.0k
Log inGet started

BlogGuides

Cline MCP server setup with Graphify, step by step

· 4 minute read

A Cline MCP server setup connects Cline, the open-source coding agent, to tools through the Model Context Protocol (MCP). Cline supports local servers that it starts as a process and remote servers it reaches at a URL. This guide adds Graphify, a code knowledge graph, both ways, so Cline can ask how your code connects instead of reading file after file.

You need Cline in VS Code or another supported editor, and a repository to try it on.

How do I add an MCP server to Cline?Link to How do I add an MCP server to Cline?

Open the Cline panel and click the MCP Servers icon. Use the Remote Servers tab to add a hosted server by URL, or open Configure, then Configure MCP Servers, to edit cline_mcp_settings.json directly. For remote servers, choose Streamable HTTP as the transport. Cline's MCP overview documents both paths.

Add hosted Graphify as a remote serverLink to Add hosted Graphify as a remote server

Connect a GitHub or GitLab repository at app.graphify.com first, and wait until it shows Indexed. Then, in the Remote Servers tab:

  1. Enter graphify as the server name.
  2. Enter https://api.graphify.com/mcp as the URL.
  3. Choose Streamable HTTP.
  4. Click Add Server.

Or add the same entry in the JSON file:

json
{
  "mcpServers": {
    "graphify": {
      "type": "streamableHttp",
      "url": "https://api.graphify.com/mcp",
      "disabled": false,
      "autoApprove": []
    }
  }
}

On first connect, your browser opens to sign in with OAuth. There is no API key to paste. The other MCP clients guide covers the hosted server for any remote client.

Why does the type field matter in Cline?Link to Why does the type field matter in Cline?

If you leave out "type": "streamableHttp", Cline treats a URL entry as the legacy SSE transport. A server that supports only Streamable HTTP then fails to connect, often with a 405 error. Write the value exactly, in camelCase. GitHub's Cline install guide calls out the same mistake.

Add a local Graphify server insteadLink to Add a local Graphify server instead

Use the local form when code must stay on your machine. Install the open-source Graphify CLI with the MCP extra and build a graph:

bash
uv tool install "graphifyy[mcp]"
graphify extract . --code-only

The graph lands in graphify-out/graph.json. A code-only build parses your code with tree-sitter and makes no model calls. Then add a stdio entry with an absolute path:

json
{
  "mcpServers": {
    "graphify": {
      "command": "graphify-mcp",
      "args": ["/absolute/path/to/repo/graphify-out/graph.json"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

The local server exposes ten tools, including query_graph, get_neighbors and shortest_path. Keep the graph current with graphify update . after changes. See the MCP server quickstart.

| | Remote (hosted Graphify) | Local (Graphify CLI) | | --- | --- | --- | | Config | type, url | command, args | | Sign-in | OAuth in the browser | None | | Graph updates | Each push to the default branch | graphify update . or git hooks | | Best for | Shared repositories on GitHub or GitLab | Private or offline projects |

How do I test the connection?Link to How do I test the connection?

Check that the server shows a green status in the MCP Servers panel and lists its tools. A red status usually means a wrong URL, a missing type field or an expired sign-in. Then start a new task and send a prompt that must call the server:

text
Use Graphify to list the repositories you can see. If the tool fails, report the error instead of guessing.

For the local server, ask Cline to call graph_stats instead. Once that works, try a real question: "What calls the function that loads config, and which files are they in?"

Should I auto-approve Graphify tools?Link to Should I auto-approve Graphify tools?

Graph tools only read, so auto-approving them saves clicks. Add tool names to autoApprove once you trust the setup. Keep approval on for servers that can write, such as ones that edit issues or databases. Our post on what a code knowledge graph is explains what the read tools return.

Try itLink to Try it

Connect a repository at app.graphify.com and add the remote entry above. For local use, run uv tool install "graphifyy[mcp]". Questions are welcome on Discord.

Using other agents too? See how to give Cursor a code knowledge graph and best codebase context tools for AI coding assistants.

Open-source Graphify is licensed under Apache 2.0. Related reading: the Cursor setup and the Claude Code setup. Sources: Cline MCP configuration, MCP specification.

FAQLink to FAQ

Where is cline_mcp_settings.json?Link to Where is cline_mcp_settings.json?

Open it from the Cline panel: MCP Servers, Configure, then Configure MCP Servers. The Cline CLI uses ~/.cline/data/settings/cline_mcp_settings.json.

Why does Cline show a 405 error for a remote server?Link to Why does Cline show a 405 error for a remote server?

The entry is probably missing "type": "streamableHttp", so Cline tried legacy SSE. Add the type and reconnect.

Does Graphify need an API key in Cline?Link to Does Graphify need an API key in Cline?

No. The hosted server uses OAuth in the browser, and the local server reads a file on your disk.

Can Cline use Graphify across several repositories?Link to Can Cline use Graphify across several repositories?

Yes. With hosted Graphify, the server can see every repository in your workspace that you have access to.

Does the local server send my code anywhere?Link to Does the local server send my code anywhere?

No. graphify-mcp reads a graph file on your machine. Cline sends what it reads to your chosen model provider.

Try Graphify now.

Try for free