FeedbackHive MCP server
FeedbackHive runs a Model Context Protocol (MCP) server, so an AI assistant can work with your projects, tasks, task groups, notes and comments directly: list what is open, create and update tasks, write notes and reply in comments, all from a chat.
Server URL: https://mcp.feedbackhive.app/api/mcp-data/mcp
Authentication
The server speaks MCP over Streamable HTTP and accepts OAuth access tokens and personal access tokens. Either way, the assistant acts as your account and sees only what your role in each project allows.
OAuth, for apps that sign you in
FeedbackHive is its own OAuth authorization server. A client discovers it from the server, then uses the authorization code flow with PKCE (S256).
- Protected resource metadata:
https://mcp.feedbackhive.app/.well-known/oauth-protected-resource - Authorization server metadata:
https://feedbackhive.app/.well-known/oauth-authorization-server - Authorization endpoint:
https://feedbackhive.app/oauth/authorize - Token endpoint:
https://feedbackhive.app/oauth/token - Token revocation (RFC 7009):
https://feedbackhive.app/oauth/revoke - Dynamic client registration (RFC 7591):
https://feedbackhive.app/oauth/register
Every code exchange also returns a refresh token, which rotates on each use, so a connected app renews its access without sending you back to sign in. It expires after 30 days without use, and one sign-in lasts at most 90 days.
A client that registers itself always shows you FeedbackHive's consent screen, marked as not verified, and can ask for at most these permissions:
- Use the FeedbackHive tools, including reading your projects, tasks, notes and comments, and running AI features (
mcp:tools) - Create, change and delete projects and their contents, including task groups, notes and their revisions, documents and folders, and use AI features that write to them (
projects:write)
On the main server, every tool that creates, changes or deletes something is refused without projects:write.
Personal access token, for everything else
Create a token in Settings, Personal Access Tokens. Choose whether it reaches all your projects or only selected ones, and when it expires. Send it as a bearer token:
Authorization: Bearer fhp_YOUR_TOKENAPI tokens (fhk_) are for the REST API and are rejected here.
Connect your assistant
Claude (claude.ai and Claude Desktop)
Add FeedbackHive as a custom connector. Claude registers itself with FeedbackHive, so there is no token to copy. Custom connectors are available on the Free, Pro, Max, Team and Enterprise plans, and the Free plan allows one.
- On Free, Pro or Max, open Customize, then Connectors, in Claude and choose Add custom connector. On Team or Enterprise, the connector is added once for the whole organization, under Organization settings, Connectors: choose Add, then Custom, then Web. That takes an Owner or Primary Owner, or on Enterprise anyone whose custom role can manage access to libraries.
- Name it FeedbackHive and paste the server URL:
Server URL https://mcp.feedbackhive.app/api/mcp-data/mcp - If Claude asks how it should identify itself, choose Register automatically. FeedbackHive supports dynamic client registration, not Claude's published identity.
- Add the connector. Then, in Customize, Connectors, choose Connect on it. On Team and Enterprise, every member does this step with their own FeedbackHive account. Claude opens a FeedbackHive sign-in, then the consent screen.
The consent screen marks Claude as not verified. That is expected: Claude registered itself, and FeedbackHive cannot confirm the name a self-registered client gives. Approve only if you started the connection yourself. Claude asks for the permissions listed under OAuth above, so it can read and change your projects, tasks, notes and comments, within what your role allows.
Claude renews its access on its own, so the connector stays signed in. You sign in again only after 30 days without using it, or 90 days after you connected it.
Connectors belong to your Claude account, so one added on claude.ai also works in Claude Desktop and the Claude mobile apps. Anthropic's help article has the details for each plan.
For the Project AI tools, add a second connector the same way with https://mcp.feedbackhive.app/api/mcp-auth/mcp. The Free plan's single custom connector is better spent on the server URL above.
ChatGPT
Add a custom MCP connector in ChatGPT's settings, paste the server URL and choose OAuth. ChatGPT registers itself with FeedbackHive, then opens a FeedbackHive sign-in and the consent screen. Approve, and the tools appear in your chats.
Custom MCP connectors are not available on every ChatGPT plan or workspace role. OpenAI's help article lists who can add one.
https://mcp.feedbackhive.app/api/mcp-data/mcpClaude Code
Add the server with its URL alone:
claude mcp add --transport http feedbackhive https://mcp.feedbackhive.app/api/mcp-data/mcpThen run /mcp in Claude Code and follow the sign-in in your browser. Claude Code registers itself, shows you the same consent screen as above and renews its access on its own.
Or add it with a personal access token instead:
claude mcp add --transport http feedbackhive https://mcp.feedbackhive.app/api/mcp-data/mcp \
--header "Authorization: Bearer fhp_YOUR_TOKEN"Cursor
Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json in a project:
{
"mcpServers": {
"feedbackhive": {
"url": "https://mcp.feedbackhive.app/api/mcp-data/mcp",
"headers": {
"Authorization": "Bearer fhp_YOUR_TOKEN"
}
}
}
}Other MCP clients
Point the client at the server URL over Streamable HTTP. If it supports MCP OAuth with dynamic client registration, it can sign you in on its own; otherwise give it the Authorization header above.
Tools
The server exposes 24 tools. They return your data as JSON and run no AI of their own. This list is read from the server code, so it matches what your client sees.
| Tool | What it does |
|---|---|
| list_projects | List projects accessible to the authenticated account with normalized metadata, including each project's task statuses (id, name, color, isDefault) in board order |
| get_project | Get normalized project metadata, its task statuses (id, name, color, isDefault) in board order, and task status counts (no AI) for a project |
| create_project | Create a project accessible to the authenticated account |
| list_tasks | List tasks for a project with filters and pagination |
| list_my_tasks | List open tasks (not done or cancelled) assigned to the authenticated account across every project it can view, with the account's id and name |
| create_task | Create a task in a project |
| get_task | Get a single task with full details, including statusName and statusColor resolved from the project's statuses |
| list_comments | List comments for a task in a project |
| create_comment | Create a comment on a task in a project |
| list_notes | List notes for a project with pagination |
| get_note | Get a single note with full details |
| create_note | Create a note in a project |
| update_note | Update a note's title, content, folder, or pinned state |
| delete_note | Permanently delete a note from a project |
| list_task_groups | List task groups for a project, ordered by position |
| create_task_group | Create a new task group in a project |
| update_task_group | Update a task group's name, icon, color, or display order |
| delete_task_group | Delete a task group |
| list_note_folders | List note folders for a project, ordered by position |
| create_note_folder | Create a new note folder in a project |
| update_note_folder | Update a note folder's name, icon, color, or display order |
| delete_note_folder | Delete a note folder |
| update_task | Update an existing task's fields |
| delete_task | Permanently delete a task from a project |
Example prompts
Once connected, ask in plain language. For example:
- List my open FeedbackHive tasks across all projects and tell me which ones are due this week.
- In the Website redesign project, create a task called "Fix the mobile menu overlap" in the Bugs group, then comment that it only happens below 375px.
- Read the notes in the Client onboarding project, summarize what is still undecided, and move every note about pricing into a Pricing folder.
Project AI endpoint
A second endpoint, https://mcp.feedbackhive.app/api/mcp-auth/mcp, serves tools that run FeedbackHive's own AI models on your projects. It accepts the same credentials.
With OAuth, its tools that create or delete tasks also need the tasks:write scope, which a self-registered client cannot hold, so connect it with a personal access token to use them.
| Tool | What it does |
|---|---|
| list_projects | List projects accessible to the authenticated account |
| get_project_summary | Get an AI-generated summary of project tasks including status breakdown, upcoming items, and suggested actions |
| get_task_focus_advice | Get AI recommendations on which tasks to focus on today based on priority, due dates, and status |
| create_task_from_description | Create a new task from a natural language description |
| classify_user_intent | Classify user message intent to determine if they want a summary or to create a task |
| delete_task | Permanently delete a task from a project |
Related
- The fbh CLI uses the same server from your terminal.
- REST API reference for integrations that do not use MCP.
- Privacy policy.
Frequently asked questions
- Which AI clients can connect to the FeedbackHive MCP server?
- Any client that speaks MCP over Streamable HTTP. Claude (claude.ai, Claude Desktop and Claude Code) and ChatGPT connect with OAuth: they register themselves and you sign in to FeedbackHive. Cursor and other clients that let you set a request header connect with a personal access token.
- What can the assistant see and change?
- The same projects your FeedbackHive account can reach, with your role's permissions. A personal access token can be limited to selected projects. On the main server, an app connected with OAuth can only create, change or delete something if its access includes the projects:write permission.
- Do I have to sign in again after connecting with OAuth?
- Rarely. Along with a short-lived access token, FeedbackHive gives the app a refresh token it uses to renew access on its own, so the connection stays signed in. You sign in again only after 30 days without using it, or 90 days after you connected it.
- How do I take access back?
- Revoke a personal access token in Settings, under Personal Access Tokens, and it stops working at once. For an app connected with OAuth, disconnect it in that app. In a Claude Team or Enterprise organization, each member disconnects their own account, since removing the connector is for the people who manage it. FeedbackHive has no screen yet to revoke an app's access from its side, and access an app stops using ends on its own after 30 days.
- Can I use my FeedbackHive API token?
- No. API tokens (starting with fhk_) are for the REST API and the MCP server rejects them. Use a personal access token (starting with fhp_) or OAuth.
- Does the MCP server run AI on my data?
- The main server returns your data as plain JSON and runs no AI of its own: your assistant does the reasoning. A second, optional endpoint offers tools that run FeedbackHive's own AI, such as project summaries.