Integrates with Stripe for subscription billing, including checkout with monthly/yearly billing intervals, payment terms, webhook handling, and subscription status checks that gate access to Milestone tools.
Integrates with Supabase for OAuth authentication and dynamic client registration, using the configured Supabase URL and anon key to sign in users and authorize MCP tool calls.
This server has a remotely hosted endpoint. Open its connector listing for connection details, authentication requirements, and setup instructions:
Milestone keeps a freelancer's approved milestone definitions and what done means, then lets an assistant read that record before it answers. An assistant cannot say a milestone is complete, or that the next payment is released, unless the saved acceptance criteria were met. It cannot invent an extra deliverable or mark work done that was not saved.
It works with ChatGPT, Claude, Gemini, Grok, and Cursor, plus any other MCP client that can do Streamable HTTP and OAuth. It is not a ChatGPT-only plugin.
Sign in with your Milestone account when the assistant opens OAuth. Do not paste an API key or password into a header. Milestone supports dynamic client registration: leave the client id and secret empty. The protected-resource metadata at /.well-known/oauth-protected-resource/mcp points clients at the OAuth issuer, which registers them.
Milestone tools need Pro or an active trial. A new subscription includes a 14-day trial. This page does not list a price. Checkout shows the billing interval and payment terms.
To self-host, run the server and use the base URL you configure. The default MCP address is http://127.0.0.1:3000/mcp.
MCP server URL: https://milestone-continuity2.vercel.app/mcp (Streamable HTTP, OAuth sign-in)
Status: early access. Paste the URL into Claude, Cursor, Grok, or ChatGPT developer mode.
Registry name: io.github.LAHutchins91/milestone
Related MCP server: Scope by Ouroboros Apps
After you approve the connection, the server exposes these tools:
list_milestone_sets
open_milestone_set
read_milestone_set
define_milestone
save_acceptance_criterion
save_deliverable
save_work_item
record_criterion_met
declare_milestone_complete
declare_next_payment_released
mark_work_done
write_client_wording
approve_milestones
suggest_milestone_change
accept_milestone_change
read_milestone_set is the read the assistant should do before it answers. It includes the milestone definitions, the acceptance criteria that say what done means, the saved deliverables and work, and what the assistant may tell the client. Draft definitions are not an approved commitment. A suggested milestone change does not change the record. After the milestones are approved, declare_milestone_complete refuses to call a milestone complete and declare_next_payment_released refuses to say the next payment is released unless every saved acceptance criterion on that milestone was met. save_deliverable refuses an extra deliverable, and mark_work_done refuses work that was not saved. Those additions go through suggest_milestone_change and then accept_milestone_change, and only when you explicitly approve that change.
The assistant only calls these tools when you and the host allow it.
Cursor, in ~/.cursor/mcp.json or a project .cursor/mcp.json:
{
"mcpServers": {
"milestone": {
"url": "http://127.0.0.1:3000/mcp"
}
}
}Do not add a headers block. Cursor registers a client and opens sign-in.
Claude Code:
claude mcp add --transport http milestone http://127.0.0.1:3000/mcpDo not pass an Authorization header. Other clients use the same address, choose OAuth, and leave client id and secret empty. Steps for ChatGPT, Claude, Gemini, Grok, and Cursor are on the connect page at /connect.
Registry metadata for this server is in server.json (io.github.LAHutchins91/milestone). The public remote MCP URL there is https://milestone-continuity2.vercel.app/mcp.
npm install
npm test
npm run typecheck
npm run build
npm startWhen stdin is a terminal, Milestone serves Streamable HTTP on port 3000. When stdin is not a terminal, it speaks MCP over stdio and still opens the HTTP port. Logs during stdio mode go to stderr so they do not mix with the protocol.
Records are stored durably in a JSON file. The default path is ~/.milestone/milestone.json. Set MILESTONE_DATA_PATH to move it. One server process owns that file. Do not point it at another product's data file.
OAuth uses the same idea as a Supabase authorization server with dynamic client registration. Set these on the server process, not in an MCP header:
APP_BASE_URL (default http://localhost:3000)
SUPABASE_URL
SUPABASE_ANON_KEY
STRIPE_SECRET_KEY
STRIPE_WEBHOOK_SECRET
STRIPE_PRICE_MONTHLY and STRIPE_PRICE_YEARLY (Stripe catalog ids, not a product price)
OPENAI_APPS_CHALLENGE (plain-text token served at /.well-known/openai-apps-challenge; the path returns 404 when unset)
Tool calls other than discovery require a signed-in account whose subscription status is active or trialing.
More from Ouroboros: https://ouroborosapps.com
Apply one suggested milestone change after the freelancer explicitly approves that change. Pass confirmed true only then. This is the path for an extra deliverable, a new acceptance criterion, or new saved work after approval. Calling it is not a substitute for the freelancer's approval.
Inputs: changeId (required), confirmed (required), milestoneSetId (required). Returns: data
Approve the draft milestone definitions and acceptance criteria. Pass confirmed true only after the freelancer explicitly approves them. A draft is not a commitment.
Inputs: confirmed (required), milestoneSetId (required). Returns: data
State that a milestone is complete. Refuses unless the milestones are approved and every saved acceptance criterion on that milestone was met. Do not tell the client a milestone is complete when this tool refuses.
Inputs: milestoneId (required), milestoneSetId (required). Returns: data
State that the next payment for a milestone is released. Refuses unless the milestones are approved, every saved acceptance criterion on that milestone was met, and the milestone is complete. Do not tell the client the next payment is released when this tool refuses.
Inputs: milestoneId (required), milestoneSetId (required). Returns: data
Add one milestone to a draft set: a title and the definition of that milestone. After milestones are approved, a new milestone is refused until accept_milestone_change applies an add_milestone suggestion.
Inputs: title (required), definition (required), milestoneSetId (required). Returns: data
List the signed-in freelancer's milestone sets. Use a returned id with read_milestone_set. Do not guess a set.
Inputs: offset. Returns: data
Mark one saved work item done. Refuses when the work item was not saved. Do not mark unsaved work done.
Inputs: workItemId (required), milestoneId (required), milestoneSetId (required). Returns: data
Open a draft milestone set for one client project. Draft definitions are not an approved commitment until approve_milestones.
Inputs: reference (required), clientName (required), projectTitle (required). Returns: data
Read the milestone set before answering. Quote only this record. mayTellClient and clientWording are what the assistant may tell the client. Do not say a milestone is complete unless complete is true. Do not say the next payment is released unless paymentReleased is true. Do not invent a deliverable. Do not mark unsaved work done. Proposed changes are not authorization.
Inputs: milestoneSetId (required). Returns: data
Record that one saved acceptance criterion was met. The criterion must already be saved. This does not by itself tell the client the milestone is complete or that the next payment is released.
Inputs: criterionId (required), milestoneId (required), milestoneSetId (required). Returns: data
Save one acceptance criterion, which is what done means for that milestone. After milestones are approved, a new criterion is refused until an accepted add_criterion suggestion. This does not mark the milestone complete.
Inputs: statement (required), milestoneId (required), milestoneSetId (required). Returns: data
Save one deliverable that belongs to a milestone. After milestones are approved, an extra deliverable is refused until an accepted add_deliverable suggestion. Do not invent a deliverable that was not saved.
Inputs: title (required), detail (required), milestoneId (required), milestoneSetId (required). Returns: data
Save one piece of work under a milestone so it can later be marked done. After milestones are approved, new work is refused until an accepted add_work_item suggestion. Unsaved work cannot be marked done.
Inputs: title (required), milestoneId (required), milestoneSetId (required). Returns: data
Record a suggested change. This does not change the milestones. kind add_milestone requires title and definition. kind add_criterion requires milestoneId and statement. kind add_deliverable requires milestoneId, title, and detail. kind add_work_item requires milestoneId and title. kind revise_client_wording requires wording.
Inputs: kind (required), summary (required), milestoneSetId (required), title, +5 more. Returns: data
Record what the assistant may tell the client about milestones, what done means, completion, and payment release. After approval, different wording is refused until an accepted revise_client_wording suggestion. The assistant must not go beyond this wording and the facts in read_milestone_set.
Inputs: wording (required), milestoneSetId (required). Returns: data
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 15 tools
The tools map to distinct workflow steps and the descriptions repeatedly clarify state constraints. The main overlap is conceptual: approve_milestones vs accept_milestone_change, and save_* vs record_*/mark_* tools require careful reading, though each has a separate purpose.
All tool names are snake_case and begin with a verb (list_, open_, read_, define_, save_, approve_, suggest_, accept_, record_, mark_, declare_, write_). The pattern is predictable throughout, including multi-word object names.
Fifteen tools are justified by the milestone lifecycle: set management, draft authoring, approval, change requests, completion declarations, payment release, and client wording. Each tool corresponds to a distinct state transition rather than redundant surface area.
Core lifecycle coverage is strong from draft creation through completion and payment release, including a suggested-change path. However, there are no update/delete tools for existing milestone definitions, criteria, deliverables, or work items, so correcting a mistaken draft entry is a notable gap.
Tasks, timers and earnings for freelancers and agencies: your AI agent works where you get paid.
Free client billing for AI assistants: hours, expenses, timers, invoices and payments.
Freelance deposit and payment schedule for AI assistants
AI invoice follow-up emails for freelancers. Draft-only, never sends for you.
