FlowDot MCP server

Read from the package source without running it. The installed server may list more.

• add_connectionwrite action

  Create a connection between two nodes in a workflow. Connects an output socket to an input socket.

• add_custom_node_comment

  Add a comment to a custom node. Can also be used to reply to existing comments.

• add_node

  Add a new node to a workflow. Supports built-in nodes, custom nodes, and toolkit-tool nodes. For custom nodes, use "custom_node_{hash}". For a tool from an INSTALLED toolkit, use "toolkit_{toolkitId}_{toolName}" (e.g. "toolkit_xS03HmGAag_get_markets") — the tool's inputs become input sockets, each required credential becomes a "_cred_<KEY>" input resolved from your installation, and the node outpu

• add_recipe_store

  Add a new store (variable) to a recipe. Stores hold data that flows between steps. **Store Types:** - **Input stores** (is_input: true): Values user provides when running - **Output stores** (is_output: true): Values returned after execution - **Internal stores**: Temporary values used during execution **IMPORTANT - The `request` Convention:** When users run a recipe with a task argument like: `fl

• add_shared_result_comment

  Add a comment to a shared execution result.

• add_workflow_comment

  Add a comment to a workflow. You can only comment on public workflows or your own workflows.

• agent_chat

  Chat with the FlowDot AI agent. The agent can help discover workflows, suggest which workflow to run, and explain workflow capabilities.

• append_app_code

  Appends content to the end of app code. CRITICAL: Remember export rules: - App entry file MUST end with: export default MyAppName; - When appending, content goes AFTER the closing brace but BEFORE the export - Use insert_app_code if you need to place code before the export statement Use this tool when you need to: - Add helper functions after the main component - Add additional components at the e

• audit_property

  Audit a FlowDot property — a deterministic, full-scope inspection of what it can DO and REACH — before recommending, installing, or running it. **What you get back** - **Capability tags** at a glance: `network`, `knowledge_base`, `llm`, `file`, `external_effect`, `toolkit`, `credentials`, `sub_invoke`, `human_gate`. - **Network calls** — the literal outbound hosts + methods it can hit. - **Knowled

• browse_community_input_presets

  Browse community-shared input presets across ALL workflows on the platform. Returns only presets the author has opted in to community visibility (`is_community=true`). This is the preset equivalent of `get_public_workflows`. To run one: read the preset's `inputs` payload, then call `execute_workflow` against the preset's parent workflow with those inputs. The parent workflow hash is returned with

• browse_recipes

  Browse public agent recipes shared by other users. Search by name, category, or tags.

• browser_act

  Perform one UI action. action ∈ click | double_click | right_click | type | clear | press_key | scroll | hover | wait | upload_file | select_option | navigate. target = element text/selector/coords; value = text/key/url/file-path.

• browser_close

  Close one browser/Electron session and free its resources.

• browser_describe

  Read the current page: title, url, headings, interactive elements (with selectors), visible text, forms, modals, and errors. The primary observation channel — use before acting.

• browser_fill_form

  Fill many form fields in one call, each with the correct interaction for its type (react-select, autocomplete, file upload, radio, checkbox). Autocomplete fields with no matching suggestion are left blank and reported failed — never filled with a wrong value.

• browser_find

  Find candidate elements matching a natural-language description or text. Returns ranked selectors.

• browser_launch

  Open a web page in a real browser, OR launch an Electron app to QA it. Returns a session_id used by every other browser_* tool. Prefers your system Chrome/Edge (no download). Requires Playwright (optional dependency); returns an install hint if unavailable. Read the `learn://browser-driving` resource first for the observe→act loop and limits.

• browser_list_sessions

  List all open browser/Electron sessions.

• browser_navigate

  Navigate a browser session to a new URL. Blocked from reaching the internal FlowDot node server (hub-mode).

• browser_read_form

  Return a schema of the page's main form: each field's label, ref, type (text|native-select|react-select|autocomplete|file|radio|checkbox|...), required flag, value, and dropdown options. Reads the TOP document only (iframed boards not covered). Use before browser_fill_form.

• browser_screenshot

  Capture a PNG of the current page. Returns a base64 image (or writes to save_path if given).

• browser_sequencewrite action

  Run several actions in one call (fast, fewer round-trips). Each item is { action, target?, value? }.

• browser_wait_for

  Wait until a selector or text appears (up to a timeout).

• cancel_decisionwrite action

  Request cancellation of one of your decision calls. Best-effort once dispatched; a finished call is only acknowledged.

• cancel_executionwrite action

  Cancel a running workflow execution. Only works on executions with status "running" or "pending".

• checkpoint_recipewrite action

  Create a manual checkpoint snapshot of a recipe. Unlike auto-snapshots (which coalesce within 60 seconds), a checkpoint is always recorded as its own version with parent_kind="checkpoint". Use before a multi-step rewrite so you can restore precisely to this point.

• clone_app

  Clone a public app to your library. Creates a private copy that you can customize. The cloned app will: - Be owned by you - Start as private (not published) - Have all the original code and configuration - Not include linked workflows (you'll need to link your own)

• convert_document

  Convert a document to another format; the output extension sets the target. Supported:

• copy_custom_node

  Copy a public custom node to your library. The copy will be private by default.

• create_agent_characterwrite action

  Create a new agent character (voice-call persona). All voice/STT/LLM/persona fields are required — the Hub rejects incomplete characters at save time with code CHARACTER_VOICE_CONFIG_INCOMPLETE. See learn://characters for recommended per-provider values.

• create_app_filewrite action

  Create a new file in a multi-file FlowDot app. Creates a new file with the specified path and content. File constraints: - Max file size: 100KB per file - Max total app size: 512KB across all files - Allowed extensions: jsx, js, tsx, ts, css, json, md - No path traversal (../) allowed - No hidden files (starting with .)

• create_custom_nodewrite action

  Create a new custom node with inputs, outputs, and script code. SCRIPT FORMAT (REQUIRED): ```javascript function processData(inputs, properties, llm) { // Access inputs by their exact names const myInput = inputs.InputName || ''; const myProp = properties.propertyKey; // Your logic here // Return object with keys matching output names EXACTLY return { OutputName: result }; } ``` VALID DATA TYPES:

• create_documentwrite action

  Create a new document at a local path; the file extension sets the format

• create_input_presetwrite action

  Create a new input preset (community input) for a workflow. This allows you to share pre-configured input values with others.

• create_knowledge_categorywrite action

  Create a new category to organize documents in your knowledge base. Categories help group related documents for targeted RAG queries. Can create personal or team categories.

• create_recipewrite action

  Create a new agent recipe. Recipes are reusable agent orchestration workflows. **After Creating:** 1. Save the returned hash - you need it for all subsequent operations 2. Add input stores (is_input: true) - **name primary input `request`** (every run surface passes the task as `inputs.request`: CLI `--input`, the native app's Run Focus, and the coding chat's `run_recipe`) 3. Add output stores (is

• create_shared_resultwrite action

  Create a shareable link for a workflow execution result. This allows you to share specific execution outputs with others.

• create_workflowwrite action

  Create a new empty workflow with a name and optional description. The workflow will be private by default.

• delete_agent_characterwrite action

  Permanently delete an agent character. Requires confirm: true. This is irreversible — any voice-call session in flight will fail with a 404 on its character lookup.

• delete_appwrite action

  Permanently delete an app. This action cannot be undone. The app and all its associated data (comments, votes, etc.) will be removed. Existing clones of the app will continue to work for their owners.

• delete_app_filewrite action

  Delete a file from a multi-file FlowDot app. Permanently removes the file from the app. This action cannot be undone. Note: Deleting the entry point file will require you to set a new entry point using set_app_entry_file.

• delete_connectionwrite action

  Remove a connection between two nodes in a workflow.

• delete_custom_nodewrite action

  Delete a custom node. This action cannot be undone.

• delete_imagewrite action

  Delete one of your uploaded images by id. Removes the stored file and frees the storage it used. Only your own images can be deleted.

• delete_input_presetwrite action

  Delete an input preset. You can only delete presets you created.

• delete_knowledge_categorywrite action

  Delete a knowledge base category. Documents in the category will become uncategorized (not deleted).

• delete_knowledge_documentwrite action

  Permanently delete a document from your knowledge base. This removes the file and all associated chunks/embeddings.

• delete_nodewrite action

  Delete a node from a workflow. This also removes all connections to/from the node.

• delete_recipewrite action

  Permanently delete an agent recipe. This action cannot be undone.

• delete_recipe_stepwrite action

  Delete a step from a recipe. References to this step will need to be updated.

• delete_recipe_storewrite action

  Delete a store from a recipe. Steps using this store will need to be updated.

• delete_workflowwrite action

  Permanently delete a workflow. This action cannot be undone. Only the workflow owner can delete it.

• document_template

  Manage saved document templates: a named spec you build from repeatedly, with per-use

• duplicate_agent_character

  Duplicate one of your own agent characters. Unlike fork, this copies the full LLM choice as well — useful for spinning up a variant.

• duplicate_workflowwrite action

  Create a copy of an existing workflow. The new workflow will be private and editable.

• edit_app_code

  Performs exact string replacement in app code. Similar to find-and-replace. CRITICAL: Remember these rules when editing code: - NO IMPORTS - React is global - MUST include export default at the end: export default MyAppName; - NO FORM ELEMENTS - Use div + button with type="button" instead - Workflow results: Use getNodeOutput(result, nodeTitle) helper - ALL BUTTONS need type="button" attribute Use

• edit_document

  Apply structured edits to a document in place and save it. Ops:

• email_archivewrite action

  Archive an email message (removes from inbox). Subject to mailbox grant permissions.

• email_deletewrite action

  Move an email message to trash. Subject to mailbox grant permissions. This is a soft delete (trash), not permanent.

• email_draftwrite action

  Create an email draft in a connected mailbox. The draft is saved but not sent.

• email_labelwrite action

  Add or remove labels on an email message. Subject to mailbox grant permissions.

• email_list_threadswrite action

  List email threads from a connected mailbox. Returns recent threads with subject, participants, and message count.

• email_readwrite action

  Read a specific email message by ID. Returns full message content including headers, body, and attachments list.

• email_replywrite action

  Reply to an email message. Subject to mailbox grant permissions.

• email_searchwrite action

  Search emails in a connected mailbox. Requires an active email integration.

• email_sendwrite action

  Send an email through a connected mailbox. Subject to mailbox grant permissions —

• execute_decisionwrite action

  Run a typed Jev decision (noul yes/no, choice, score) and wait for the result. SPENDS the user's credits

• execute_workflowwrite action

  Execute a FlowDot workflow by ID. Use list_workflows first to get available workflow IDs.

• get_decision

  Read one of your decision calls by call_id (status, settlement, answers when available). Never creates work.

• get_document_info

  Get a document's structured outline: format, per-format summary (counts of slides/pages/sheets/paragraphs),

• inspect_document

  File-level PDF detail an outline cannot carry: byte size on disk, page count, every

• learn_about

  Load a FlowDot area: returns a short how-to AND unlocks that area's full

• list_decision_models

  List decision models with primitives, limits, price per million input tokens, availability and whether this connection may execute/read/cancel.

• list_my_resources

  List the current user's existing FlowDot resources of a given type (so you can

• mcp__flowdot__add_toolkit_comment

  Add a comment to a toolkit. Share feedback, ask questions, or discuss the toolkit with the community.

• mcp__flowdot__check_toolkit_credentials

  Check credential status for a toolkit installation. Shows which credentials are required and which are missing.

• mcp__flowdot__copy_agent_toolkit

  Copy/duplicate a toolkit. Creates a private copy of a toolkit that you can customize.

• mcp__flowdot__create_toolkit_toolwrite action

  Create a new tool in a toolkit. Add HTTP or Workflow-based tools to your agent toolkit with custom input/output schemas and credentials. For signed-request crypto exchange APIs (Robinhood Crypto, Binance, etc.), use endpoint_config.signing to declare the algorithm and message template — the Hub computes a fresh signature on every request.

• mcp__flowdot__delete_agent_toolkitwrite action

  Delete a toolkit permanently. WARNING: This cannot be undone. All tools, installations, and related data will be deleted.

• mcp__flowdot__delete_toolkit_toolwrite action

  Delete a tool from a toolkit. WARNING: This cannot be undone. The tool will be permanently removed.

• mcp__flowdot__favorite_toolkit

  Add or remove a toolkit from your favorites.

• mcp__flowdot__get_agent_toolkit

  Get detailed information about a specific agent toolkit. Returns comprehensive toolkit details including tools, credentials, and metadata.

• mcp__flowdot__get_toolkit_comments

  Get comments on a toolkit. View community feedback and discussions about a toolkit.

• mcp__flowdot__get_toolkit_tool

  Get detailed information about a specific tool in a toolkit. Shows input schema, configuration, and requirements.

• mcp__flowdot__install_toolkit

  Install a toolkit to your account. After installation, you can configure credentials and use the toolkit's tools.

• mcp__flowdot__invoke_toolkit_toolwrite action

  Execute a tool from an installed toolkit. Runs the tool with provided inputs using your configured credentials. You can optionally pass credential_overrides to provide dynamic credentials (like tokens from a previous tool call).

• mcp__flowdot__list_agent_toolkits

  List your own MCP agent toolkits with optional filtering. Toolkits are collections of custom tools that can be installed and used by AI agents. Each toolkit can contain multiple related tools with shared credentials and configuration.

• mcp__flowdot__list_installed_toolkits

  List all toolkits you have installed. Shows installation status and credential configuration for each toolkit.

• mcp__flowdot__list_toolkit_tools

  List all tools in a toolkit. Shows tool names, descriptions, and types (HTTP or Workflow).

• mcp__flowdot__search_agent_toolkits

  Search the public MCP toolkit marketplace for toolkits created by other users. Discover reusable tool collections for various use cases like API integrations, data processing, automation, and more.

• mcp__flowdot__test_toolkit_installationwrite action

  Run a Test Connection diagnostic against an installed toolkit. Resolves the toolkit's health-check tool (the one designated by `health_check_tool_id`, or auto-detected as the first GET-method HTTP tool with no required inputs) and invokes it with empty inputs. Returns a structured diagnostic outcome. Use this to verify a toolkit is actually working before running production tools. Possible `test_o

• mcp__flowdot__toggle_toolkit_active

  Enable or disable a toolkit installation. Inactive toolkits won't be available for use but remain installed.

• mcp__flowdot__toggle_toolkit_visibility

  Change toolkit visibility (private, public, or unlisted). Public toolkits are searchable by others. Unlisted are accessible via direct link only.

• mcp__flowdot__uninstall_toolkit

  Uninstall a toolkit from your account. Removes the installation and credential mapping (toolkit itself remains available).

• mcp__flowdot__update_agent_toolkitwrite action

  Update an existing agent toolkit's metadata. Modify title, description, category, tags, and other toolkit properties. ## Updating the public README Use `readme` to set the GitHub-style markdown README shown on the public toolkit page (`/toolkit/{hash}`). Only the toolkit's author (or a team member with edit rights) can change it. Pass an empty string to clear it. Max 100,000 characters. ## Updatin

• mcp__flowdot__update_toolkit_installationwrite action

  Update toolkit installation (configure credential mapping). Map toolkit credential requirements to your stored API keys.

• mcp__flowdot__update_toolkit_toolwrite action

  Update an existing tool in a toolkit. Modify tool configuration, schemas, endpoints, or credentials.

• mcp__flowdot__vote_toolkit

  Vote on a toolkit (upvote, downvote, or remove your vote).

• merge_documents

  Concatenate PDFs into a single file, in the order given. Page content only —

• read_document

  Read the full text of a local document (pdf, docx, pptx, xlsx, md, txt, csv) by file path.

• Code scan388 source files scanned20/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitystatic API keys via environment variables6/15
• Maintenancelast push 10 days ago15/15
• Maintainer identitynamespace and repository owner differ; website matches verified namespace6/10

Safety score

72/100

higher than 38% of 48,255 servers

Tools

100

the typical server has 7

Tools that change things

48%

typical: 0%. Fewer write tools means less an agent can break

Downloads a week

68

more than 53% of 12,191 servers with a package

FlowDot: common questions

Is FlowDot MCP server safe?

Mostly: it is graded B (72/100). Read the FlowDot safety report

How do I install FlowDot?

It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.

Does FlowDot need an API key?

Yes. The registry entry asks for FLOWDOT_API_TOKEN.

Is FlowDot maintained?

The last commit was 31 days ago (2026-10-04). The latest release is v1.3.21.

How does FlowDot compare with other MCP servers?

safety score 72/100 (higher than 38% of 48,255 servers); tools 100 (the typical server has 7); tools that change things 48% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.