Move from local to Cloud

Move a repository from the local graphifyy package to Graphify Cloud. Cloud indexes the repository on GitHub or GitLab. Your assistants then query that graph through the hosted server at https://api.graphify.com/mcp. This is a configuration change. Your application code stays as it is. Each assistant keeps reading the local graph until you switch its graphify entry. Leave graphify-out/ in place until you are sure the hosted server answers, so you can roll back.

Overview

Scope	Effort	Downtime
Assistant configuration, tool calls, and saved lessons	Configuration only. Your application code stays as it is.	None. Each assistant reads the local graph until you switch it.

Cloud indexes repositories on GitHub or GitLab. A project that exists only on your machine stays on the local package.

Why move to Cloud?

The graph updates itself. Cloud indexes on each push to the default branch. You stop running graphify update . and the git hooks that rebuild the local graph.
One connection for the workspace. Assistants sign in once and name any indexed repository, such as your-org/your-repo. Local paths leave the MCP config.
Reviews on every pull request. Graphify comments when a pull request is opened, reopened, or marked ready for review.
Learning from your team’s replies. Feedback on findings tunes what later reviews surface.
Formal verification on changed functions. Each result says whether behavior is preserved, or which input produces a different result.
A wiki you can search and ask. Generated from the indexed graph, with links back to the code in Graphs.
Shared memory for the team. Notes for a repository are available to every assistant and teammate in the workspace.
OAuth sign-in. Each assistant signs in through the browser. There is no API key to paste into a config file.

Plan

Sign in. Open app.graphify.com and sign in with Google, GitHub, GitLab, or email. If you are new to Graphify, choose Create an account on the login page.
Check the repository. It is on GitHub or GitLab, and the commits you want in the graph are on its default branch. You need permission on the provider to grant Graphify access to it.
Choose how assistants sign in. Each assistant signs in with OAuth the first time it connects. An agent that cannot open a browser, such as one in CI, uses an API token instead.
Find every local use. From the repository root, run the list command for each assistant you use:

claude mcp list
codex mcp list
graphify hook status
git grep --untracked -n -e graphify-mcp -e graphify-out -e "graphify query" -e "graphify extract"

Look for MCP entries named graphify that run graphify-mcp, skill registrations from graphify install, git hooks, CI jobs that run graphify extract, and scripts that read graphify-out/. Also search your assistant rules for local tool names such as get_node and god_nodes. In Cursor and VS Code, open the MCP files named under each assistant in Switch each assistant to the hosted server.

Move with your assistant

Paste this prompt into the assistant you are moving. It reads this page, lists every local Graphify piece it finds, and shows the change it proposes for each. After you approve that list, it makes the edits and stops so you can sign in yourself.

Help me move this repository from the local Graphify package (graphifyy) to Graphify Cloud.

Read https://docs.graphify.com/platform/local-to-cloud.md first and follow it for this assistant.

Start with a plan. List every MCP server entry named graphify that runs graphify-mcp, in each scope and configuration file this assistant reads. Also list Graphify skill files, git hooks, and any rule, prompt, or script that names a local Graphify tool. For each item, show the change you propose. Wait for my approval before you edit anything.

When I approve, replace each local graphify entry with the hosted Streamable HTTP server at https://api.graphify.com/mcp under the same name. Keep every other MCP server and setting as it is. Let me complete the OAuth sign-in myself.

Leave graphify-out/ and the local Graphify skill in place. I will remove them after the hosted server answers.

To verify, use Graphify's list_repositories tool and show the repositories I can access. If the tool fails, report the error instead of guessing. Do not claim the move succeeded until a hosted tool has answered.

Move by hand

1. Connect the repository

Open Repositories in Graphify Cloud.
Choose Connect GitHub or Connect GitLab.
Complete the provider’s repository-access flow.
Wait until the repository’s status is Indexed.

Status	What to do
Indexed	Go to step 2.
Queued or Building	Wait. The local graph keeps answering in the meantime.
Failed	Open Builds and choose Retry. Retry indexes the current head of the default branch.
Paused	Open Options for the repository and switch Indexing on.

The local graph describes your folder as it was at the last build or update, uncommitted changes included. Cloud’s graph describes the latest indexed commit on the default branch. A push to another branch, or a pull request, does not change it. See Repositories and builds.

2. Switch each assistant to the hosted server

Switch one assistant at a time. Each tab shows the local entry you have now and the hosted entry that replaces it. Keep the name graphify. Leave every other MCP server as it is. Keep the graphifyy package until the move is done. The rollback steps use it.

Claude Code checks the local scope, then the project scope, then the user scope, and uses the first graphify entry it finds. A local-scope entry for the local server hides the hosted one. Remove it first.

claude mcp add --transport stdio graphify -- graphify-mcp /absolute/path/to/repo/graphify-out/graph.json

claude mcp remove graphify --scope local
claude mcp add --transport http --scope user graphify https://api.graphify.com/mcp

Run the commands from the repository folder where you added the local server. If you added it with --scope user, run claude mcp remove graphify --scope user instead of the first command.In Claude Code, run /mcp. Select graphify and choose Authenticate. Finish the sign-in in the browser window that opens. claude mcp list then shows graphify with the URL https://api.graphify.com/mcp.If the repository commits a .mcp.json that runs graphify-mcp, replace its graphify entry:

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

{
  "mcpServers": {
    "graphify": {
      "type": "http",
      "url": "https://api.graphify.com/mcp"
    }
  }
}

Each teammate approves the shared entry once, then signs in from /mcp.

The entry is in .cursor/mcp.json in the repository, or in ~/.cursor/mcp.json for your user.

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

{
  "mcpServers": {
    "graphify": {
      "url": "https://api.graphify.com/mcp"
    }
  }
}

Keep the other servers in the file. Cursor shows Needs login until you finish the one-time sign-in in your browser.

codex mcp add and codex mcp remove edit ~/.codex/config.toml. A project can also define graphify in .codex/config.toml.

[mcp_servers.graphify]
command = "graphify-mcp"
args = ["/absolute/path/to/repo/graphify-out/graph.json"]

[mcp_servers.graphify]
url = "https://api.graphify.com/mcp"
auth = "oauth"

To make the same change from the terminal:

codex mcp remove graphify
codex mcp add graphify --url https://api.graphify.com/mcp
codex mcp login graphify

Complete the browser sign-in that codex mcp login graphify opens. A hand edit to the TOML needs that login command too. Then open a new Codex session and use /mcp to check that graphify is available.

The entry is in .vscode/mcp.json, or in the file that MCP: Open User Configuration opens.

{
  "servers": {
    "graphify": {
      "type": "stdio",
      "command": "graphify-mcp",
      "args": ["/absolute/path/to/repo/graphify-out/graph.json"]
    }
  }
}

{
  "servers": {
    "graphify": {
      "type": "http",
      "url": "https://api.graphify.com/mcp"
    }
  }
}

To skip the hand edit, remove the local entry and install the Graphify MCP extension. Use one path or the other. With both, the same tools appear twice.To sign in, run MCP: List Servers from the Command Palette. Select Graphify and start it. Your browser opens for the sign-in.

If you added the local server to ~/.copilot/mcp-config.json, remove its graphify entry. Then add the hosted server:

copilot mcp add --transport http graphify https://api.graphify.com/mcp

{
  "mcpServers": {
    "graphify": {
      "type": "http",
      "url": "https://api.graphify.com/mcp",
      "tools": ["*"]
    }
  }
}

tools is Copilot’s own list of allowed tools. Without it, the server connects and its tools stay unavailable. Copilot CLI signs in on first connect. /mcp show lists what is connected.

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

{
  "mcpServers": {
    "graphify": {
      "url": "https://api.graphify.com/mcp"
    }
  }
}

Field names vary by client. Windsurf names the URL field serverUrl. The file or command for each client is on the Graphify MCP connection guide. JetBrains, ChatGPT, OpenClaw, and Vercel have their own pages in Choose your coding assistant.

An assistant that cannot open a browser, such as one in CI, uses an API token. On the MCP tab of Connectors, choose Use an API token for CI or headless agents.

3. Check that the hosted server answers

Start a new session in the assistant and paste the connection test prompt from the console:

Use Graphify's list_repositories tool and show the repositories I can access. If the tool fails, report the error instead of guessing.

The answer lists each repository’s full name, such as your-org/your-repo, and whether its index is ready. The hosted server is connected when the assistant’s tools include list_workspaces and list_repositories. If the tools include get_node or god_nodes, the assistant still reads the local graph. Return to step 2 for that assistant. Then compare one answer. Ask a question with the local CLI:

graphify query "authentication flow"

Ask the assistant the same question about your-org/your-repo. Cloud indexes the default branch, so answers can differ where your working copy has changes the default branch lacks. Open the cited source lines before you trust an answer. To set the repository an assistant uses when a tool call names none, choose a Default repository on the MCP tab of Connectors.

4. Update tool calls

Hosted calls name the repository. One hosted connection serves every indexed repository in your workspace. Pass repository_id on each repository tool, as owner/name or the UUID that list_repositories returns. You can leave it out when the workspace has one repository.

Several inputs have new names. label is symbol. On query_graph, token_budget is budget. shortest_path takes src and tgt in place of source and target. The local project_path input becomes repository_id.

If a rule, prompt, or script calls a local tool, change it to the hosted tool for the same job.

Job	Local call	Hosted call
Ask a question	`query_graph` with `question`	`query_graph` with `repository_id` and `question`
Read one symbol	`get_node` with `label`	`graphify_node` with `repository_id` and `symbol`
List direct neighbors	`get_neighbors` with `label`	`graphify_node` for every neighbor. `graphify_callers`, `graphify_callees`, or `graphify_references` for one kind of edge.
Connect two symbols	`shortest_path` with `source` and `target`	`shortest_path` with `src` and `tgt`. `graphify_trace` for directed call paths.
Summarize the graph	`graph_stats` with no input	`graph_stats` with `repository_id`
Find the most connected code	`god_nodes`	Open Graphs in the console. It lists the most connected areas.
Read one community	`get_community` with `community_id`	Open Graphs in the console. All code areas picks an area.
Check pull requests	`list_prs`, `get_pr_impact`, or `triage_prs`	Open Reviews in the console. `graphify_impact` and `impact_and_risk` show what a change reaches.

query_graph, shortest_path, and graph_stats keep their names. Replace your-org/your-repo in these examples with a full name from list_repositories.

Ask a question

{"question": "request validation", "mode": "bfs", "token_budget": 1500}

{"repository_id": "your-org/your-repo", "question": "Where is request validation handled?", "budget": 1500}

The hosted tool has no mode, depth, or context_filter input. Full sentences work better than keywords. Set skeleton to true to get signatures and outlines in place of full definition bodies.

Read one symbol

{"label": "normalize_email"}

{"repository_id": "your-org/your-repo", "symbol": "normalize_email"}

graphify_node returns the definition and its immediate neighbors. If the name does not match, it can return a similar symbol. Check the resolved field: exact is the symbol you asked for, and semantic is a suggested match.

List the callers of a symbol

{"label": "normalize_email", "relation_filter": "calls"}

{"repository_id": "your-org/your-repo", "symbol": "normalize_email"}

Use graphify_callees for what the symbol calls, and graphify_references for reads and mentions that are not calls. strict_calls is true by default, which keeps only resolved call edges. Set it to false to include inferred edges.

Connect two symbols

{"source": "login", "target": "normalize_email", "max_hops": 8}

{"repository_id": "your-org/your-repo", "src": "login", "tgt": "normalize_email"}

{"repository_id": "your-org/your-repo", "src": "login", "tgt": "normalize_email", "k_paths": 3}

The hosted tools have no max_hops or undirected input. graphify_trace returns directed call paths, and k_paths sets how many.

Summarize the graph

{}

{"repository_id": "your-org/your-repo"}

The hosted result has node and edge counts and languages. It is also a quick check that the repository is indexed.

The hosted server also has tools that the local server lacks:

graphify_find locates where a symbol is defined, from a fragment of its name.
graphify_tests_for finds tests linked to a symbol or file.
graphify_file_neighbors and graphify_imports_exports show what a file imports and what imports it.
graphify_rank_files ranks source files by relevance to a question.
impact_and_risk ranks the files a change reaches, using their linked tests.
list_workspaces and set_workspace list and switch workspaces.
remember, recall, and memories_about save and read workspace notes.

Some tools save notes or change the selected workspace. Approve a tool only when that is what you meant. ChatGPT, Codex, and Dots describes each hosted tool.

5. Move saved lessons into Memory

If you used graphify save-result and graphify reflect, your lessons live in graphify-out/reflections/LESSONS.md on your machine. Cloud keeps its own notes for each repository in Memory. There is no import for that file. Copy the lessons you still trust by hand:

Check each lesson against the current source. Leave out any lesson the code no longer supports.
Open Memory in the console and choose the repository.
Choose Add memory and save the lesson as a note.

An assistant on the hosted server can save the same note with remember:

graphify save-result --question "Where is email normalized?" \
  --answer "normalize_email handles normalization before lookup." \
  --nodes normalize_email --outcome useful

{"repository_id": "your-org/your-repo", "text": "normalize_email handles normalization before lookup.", "tags": ["auth"]}

remember has no outcome field. Put the outcome in the text or in tags. Notes from remember can wait under Needs review in Memory until an authorized member accepts them. To read notes back, use memories_about with a file or symbol, or recall with a question:

{"repository_id": "your-org/your-repo", "target": "normalize_email"}

{"repository_id": "your-org/your-repo", "query": "How is email normalized?"}

Notes in Memory are shared with the workspace. Move only lessons that everyone in the workspace may read.

6. Retire graph builds in CI

If a CI job runs graphify extract to build a graph for assistants, remove the job once every assistant uses the hosted server. Cloud builds the graph on each push to the default branch. Keep the job if something other than an assistant reads its output. If your workspace uses Buildkite with Graphify, Cloud can also select the tests a pull request affects. See Buildkite test selection.

7. Remove the local setup

The local skill installs instructions that tell each assistant to query the local graph. Some assistants also get hooks that do the same. Remove them once every assistant answers from the hosted server. For a personal install, run this from the repository root:

graphify uninstall

That command removes your registrations for every assistant and the repository’s Graphify git hooks. It also edits instruction files in the repository, such as CLAUDE.md and AGENTS.md, and deletes .cursor/rules/graphify.mdc. Review those changes with git status before you commit. The command keeps graphify-out/ unless you add --purge. For a registration the repository commits, run this once for each assistant, then commit the change:

graphify uninstall --project --platform cursor

Teammates who installed the skill or the git hooks themselves run graphify uninstall in their own clone. When you no longer need a way back, run graphify uninstall --purge to delete graphify-out/. If the repository commits graphify-out/, commit the deletion. Then remove the package:

uv tool uninstall graphifyy

If you installed with pipx, run pipx uninstall graphifyy.

What Cloud adds

These features run on the repositories you connect. Open each one after the hosted server answers.

Pull request reviews

Why it matters: Each pull request is reviewed against the indexed graph and the pull request’s diff. You see the code a change can affect before it merges.

Graphify reviews a pull request when it is opened, reopened, or marked ready for review. To review the current head after new commits, a repository owner, member, or collaborator comments on the pull request:

@graphify review

Read the results in Reviews. See Pull request reviews.

Learning from review feedback

Why it matters: Your team’s feedback on findings tunes what later advisory reviews surface and what stays quiet.

Reply in the pull request conversation with the exact finding name from the Graphify review:

@graphify wrong <finding>
@graphify good <finding>
@graphify suppress <rule>

Learning leaves formal verification unchanged. See Learning.

Formal verification

Why it matters: A changed function is compared with its earlier implementation. The result says whether behavior is preserved, or which input produces a different result.

Open Formal Verification, choose the repository, and expand a run. A function’s result is Proved equivalent (bounded), Behavior change, No divergence sampled (not proven), Abstained, or Errored. See Formal verification.

Wiki

Why it matters: graphify export wiki writes Markdown files on your machine. The hosted wiki is generated for each indexed repository. You can search it or ask it questions in the console.

Open Wiki and select a repository. Start with Overview. Search finds words in articles. Ask answers questions that need connected code context, such as a call path or where a behavior lives. Links open the matching code in Graphs. See Wiki.

Shared memory

Why it matters: A note saved for a repository is available to every assistant and teammate in the workspace, with its source and history.

Ask what the workspace remembers about a file or symbol before you change it:

memories_about

{"repository_id": "your-org/your-repo", "target": "app/models/message.rb"}

See Agent memory.

Tests linked to a symbol

Why it matters: Before you change a symbol, you can list the tests linked to it. impact_and_risk combines the code a symbol reaches with its linked tests into a ranked list of files.

{"repository_id": "your-org/your-repo", "target": "normalize_email"}

{"repository_id": "your-org/your-repo", "target": "normalize_email"}

Both results come from the indexed graph. They do not run tests or replace a review.

Summary of changes

Area	Local package	Graphify Cloud	Action required
Graph build	`/graphify .` or `graphify extract` on your machine	Indexed from the default branch on GitHub or GitLab	Connect the repository
Updates	`graphify update .`, `graphify watch .`, or git hooks	Each push to the default branch	Remove the hooks after the move
Connection	`graphify-mcp` over stdio, with a path to `graph.json`	`https://api.graphify.com/mcp` over Streamable HTTP	Replace the `graphify` entry
Sign-in	None	OAuth in the browser, or an API token for CI	Sign in once in each assistant
Repository	One `graph.json` for each entry, or `project_path`	`repository_id` on each call	Add `repository_id` to tool calls
Tool names	`get_node`, `get_neighbors`, and `shortest_path` with `source` and `target`	`graphify_node`, `graphify_callers`, and `shortest_path` with `src` and `tgt`	Update rules and scripts
Pull requests	`list_prs`, `get_pr_impact`, and `triage_prs` through `gh`	Reviews	Read reviews in Reviews
Saved lessons	`graphify-out/reflections/LESSONS.md`	Memory for each repository	Copy lessons by hand
Change impact	`graphify affected "<node>"`	`graphify_impact` and `impact_and_risk`	Update scripts and prompts
Wiki	`graphify export wiki`	Wiki for each indexed repository	None
Several repositories	`graphify merge-graphs` or `graphify global add`	Cross-Repository in Graphs	Connect each repository
Team	Each clone installs the skill and the hooks	Members in the user menu	Invite teammates from Members

If something fails

What you see	What to do
The tools include `get_node` or `god_nodes`	The assistant still reads the local graph. Remove the local `graphify` entry in step 2. In Claude Code, `claude mcp list` shows every entry.
The same tools appear twice	Two entries reach the hosted server, such as the VS Code extension and a manual entry. Keep one.
The assistant asks for an API key	The hosted server uses OAuth. Finish the sign-in in the browser. For CI, use an API token from the MCP tab of Connectors.
A tool call fails and names repositories	The call needs `repository_id`. Pass one of the names in the error, or run `list_repositories`.
`graphify_node` returns a different symbol	The name did not match, so the tool returned a similar one. Check `resolved`, then search with `graphify_find` and a fragment of the name.
The repository is missing	It is not indexed yet, or it is in another workspace. Check its status on Repositories, and check the workspace with `list_workspaces`.
An answer misses code you can see locally	Cloud indexes the default branch. Push that code to the default branch and wait for Indexed.

Roll back

The local graph stays in graphify-out/ until you delete it, so you can go back at any point.

Restore the entry. In each assistant, replace the hosted graphify entry with the Local (before) entry from step 2. Connect the graph MCP server has the command for each client.
Reinstall what you removed. If you removed the package, run uv tool install "graphifyy[mcp]". If you removed the skill, register it again, for example with graphify install --platform cursor. Add --project for a registration the repository commits.
Refresh the graph. Run graphify update .. If you used git hooks, run graphify hook install.
Verify. Ask the assistant to call graph_stats, then query_graph with a question about the repository.

To stop Cloud from indexing the repository, open Options for it on Repositories and switch Indexing off. Its status changes to Paused.

Next steps

Graphify Cloud: open the repository you just connected, then try Reviews, Wiki, or Memory.
Pull request reviews: when a review runs and how to read its findings.
Agents and integrations: connect more assistants to the hosted server.

Console tour

Every screen in the Cloud workspace, and what each one is for.

When something fails

Fix an empty graph, a failed build, or a stuck index.

Suggest edits Raise issue

Tasks

Explore

Review

People

Workspace

Move from local to Cloud

Overview

Plan

Move with your assistant

Move by hand

1. Connect the repository

2. Switch each assistant to the hosted server

3. Check that the hosted server answers

4. Update tool calls

5. Move saved lessons into Memory

6. Retire graph builds in CI

7. Remove the local setup

What Cloud adds

Summary of changes

If something fails

Roll back

Next steps

Console tour

When something fails

​Overview

​Plan

​Move with your assistant

​Move by hand

​1. Connect the repository

​2. Switch each assistant to the hosted server

​3. Check that the hosted server answers

​4. Update tool calls

​5. Move saved lessons into Memory

​6. Retire graph builds in CI

​7. Remove the local setup

​What Cloud adds

​Summary of changes

​If something fails

​Roll back

​Next steps

Console tour

When something fails

Overview

Plan

Move with your assistant

Move by hand

1. Connect the repository

2. Switch each assistant to the hosted server

3. Check that the hosted server answers

4. Update tool calls

5. Move saved lessons into Memory

6. Retire graph builds in CI

7. Remove the local setup

What Cloud adds

Summary of changes

If something fails

Roll back

Next steps