What this is

Most project management tools were built for humans clicking through a UI. When an AI agent works on your project, it has no durable place to put what it learned — the task list records what is being done, and everything about why is lost when the chat window closes.

FrameOn is a project management platform where that context is a first-class citizen: tasks carry hierarchy, hours and cost; a wiki carries decisions with version history; and a separate project memory carries the constraints and conventions an agent should know before it touches anything. The MCP server exposes all of it to your assistant, scoped to the projects your credential can actually reach.

The server is first-party — built and operated by the FrameOn team, not a community wrapper over a public API.

Connection details

Everything a client or a directory needs in order to connect.
Endpoint	`https://api.frameonlab.com/api/v1/mcp`
Transport	Streamable HTTP (JSON-RPC 2.0)
Authentication	OAuth 2.1 — Dynamic Client Registration (RFC 7591) with PKCE `S256`. Personal access tokens are supported for clients that do not speak OAuth.
Protocol versions	`2025-03-26`, `2025-06-18`, `2025-11-25`, `2026-03-26`, `2026-06-18`
Capabilities	tools, resources, prompts
Tools	20 — 13 read-only, 7 write
Registry	`com.frameonlab/frameon` on the official MCP Registry
Pricing	MCP access is enabled by default on new workspaces, including the free tier.

Claude Code

claude mcp add --transport http frameon https://api.frameonlab.com/api/v1/mcp

Claude Desktop, ChatGPT, Cursor and other clients

Add a remote MCP server pointing at the endpoint above. The client discovers the authorization server on its own and opens a browser for consent — you do not need to paste a token.

{
  "mcpServers": {
    "frameon": {
      "type": "http",
      "url": "https://api.frameonlab.com/api/v1/mcp"
    }
  }
}

If your client does not support OAuth, generate a personal access token in FrameOn under Settings → Integrations and send it as Authorization: Bearer <YOUR_TOKEN>. The token inherits your own permissions — it never widens them.

The 20 tools

Every tool takes a project_id and is checked against the caller's role and project access before it runs. Write tools are exposed only when the credential carries the write scope.

Tools exposed by the FrameOn MCP server, as served in production.
Tool	Access	What it does
`list_tasks`	read	List tasks in a project, filtered by status, assignee, type or dates.
`get_task`	read	Full detail of a single task, including description, dates and comments.
`get_task_stats`	read	Aggregate counts by status, assignee and priority for a project.
`get_project_structure`	read	The project's task tree — epics, stories, tasks and subtasks.
`create_task`	write	Create a task anywhere in the hierarchy, with dates, estimate and labels.
`update_task`	write	Change status, assignee, dates, estimate, priority or labels of a task.
`add_comment`	write	Append a comment to a task — the running log of what happened and why.
`log_time`	write	Log hours worked against a task, with date and description.
`list_documents_brief`	read	Titles and identifiers of the project's wiki pages and files.
`get_document_full_content`	read	The full content of one wiki page or document.
`search_documents`	read	Semantic search across the project's documentation.
`create_wiki_page`	write	Create a wiki page — where a decision that outlives a task belongs.
`get_project_memory`	read	The project's durable memory: decisions, constraints and conventions an agent should know before acting.
`save_classified_memory`	write	Write a classified memory entry so the next session starts where this one ended.
`get_project_team`	read	Members of the project and their roles, for assigning work correctly.
`get_time_report`	read	Hours by person, task and period.
`get_project_alerts`	read	Open risks and alerts raised on the project — overdue work, blocked tasks.
`list_skills`	read	Skills available to agents working on this workspace.
`get_skill`	read	The full instructions of one skill.
`report_ai_usage`	write	Report token usage and cost of an AI session against a task, so the project's real cost includes what the agents spent.

Security and data access

OAuth 2.1 with PKCE S256, mandatory. Redirect URIs are matched by exact equality; http: is accepted only on loopback.
Refresh token rotation with reuse detection — a replayed token invalidates the whole family.
Scoped to the caller. A credential reaches exactly the projects its owner's role reaches. There is no tool that widens access.
Workspace isolation. Every query is filtered by workspace, and the workspace identifier is never returned in a tool response.
Nothing destructive is exposed. There is no delete tool. The write tools create records or change fields on records the caller already controls.
Rate limited per credential, with a concurrency ceiling per instance.

Terms, privacy policy and acceptable use are published at /terms, /privacy and /aup.

When FrameOn is not the right fit

Jira, Linear, Asana, ClickUp and Notion all ship MCP servers too. FrameOn is worth evaluating when the shared context — decisions, memory, and the real cost of the work including what the agents spent — matters as much as the task list. It is honestly the wrong choice when:

your team is not moving off its current tracker, and you only want an MCP layer on top of it;
you need a mature third-party integration marketplace — FrameOn is a younger product with a smaller ecosystem;
you need native desktop or mobile applications rather than the web app.

FrameOn imports from Jira and Microsoft Project, so evaluating it does not require abandoning your history.

Getting started

Create a workspace — self-service, free tier, no sales call. MCP is enabled by default.
Add the endpoint to your MCP client and approve the OAuth prompt.
Ask your assistant something like "what's open on this project and who owns it?" to confirm the connection.

Step-by-step instructions per client, in five languages, live at app.frameonlab.com/connect.