Dev Project Coordinator MCP server
Commit Nov 10, 2026Tracks coding and non-coding projects as local JSON cards that an AI agent reads and updates between sessions.
Also known as M-Pineapple/Dev-Project-Coordinator
- Stars
- 47
- Tools
- 21
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 10, 2026
- License
- Apache-2.0
What is Dev Project Coordinator MCP?
Dev Project Coordinator tracks coding and non-coding projects as local JSON cards that an AI agent reads and updates between sessions. It exposes 21 tools, such as get_desk_summary, list_projects, get_project_status.
Summary generated from the project's README
About Dev Project Coordinator MCP server
Managing coding and non-coding tasks through an AI agent is possible with Dev-Project-Coordinator. This local MCP server allows agents to track work using JSON cards stored directly on your machine, ensuring context survives between different chat sessions. It is designed for developers and professionals who mix technical tasks with writing, legal, or accounting work and want a single source of truth that does not require a cloud account or external telemetry.
Users interact with the server through tools like get_desk_summary to retrieve a compact state of all projects or register_from_workspace to automatically detect a repository and its language. The system supports a Kanban workflow with columns such as backlog, doing, review, and done. While it does not replace RAG or memory tools, it complements them by holding the live state of what is being done and where, while other tools handle the reasoning behind decisions. You can also view your progress through a local Analytics and Kanban tab hosted on your own disk.
What you can do with it
- 1Resume work between sessionsAn agent can quickly catch up on a project's status by reading a compact summary at the start of a new chat.
- 2Track non-coding professional tasksYou can organize administrative or legal work into projects that do not require a specific file path.
- 3Automate repository registrationThe server can identify a coding project from your current directory and set up a starting task card.
Written from the project's README. Every use case is checked against a quote from it.
Dev Project Coordinator MCP tools 21
Detected from the READMEget_desk_summarylist_projectsget_project_statusadd_projectupdate_project_statussearch_desksearch_code_patternsget_project_timelineget_activity_heatmapget_technology_trendsget_project_healthopen_desk
| get_desk_summary | — | |
| list_projects | — | |
| get_project_status | — | |
| add_project | — | |
| update_project_status | — | |
| search_desk | — | |
| search_code_patterns | — | |
| get_project_timeline | — | |
| get_activity_heatmap | — | |
| get_technology_trends | — | |
| get_project_health | — | |
| open_desk | — |
Show all 21 toolsShow fewer tools
add_cardmove_cardupdate_cardrename_columnregister_from_workspacefile_under_projectrename_projectmerge_projectsstart_desk_server
| add_card | — | |
| move_card | — | |
| update_card | — | |
| rename_column | — | |
| register_from_workspace | — | |
| file_under_project | — | |
| rename_project | — | |
| merge_projects | — | |
| start_desk_server | — |
Automated checks
Ran Nov 13, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablem-pineapple/dev-project-coordinator is public and reachablegithub.com/m-pineapple/dev-project-coordinator ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)Apache-2.0spdx.org/licenses/Apache-2.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-10github.com/m-pineapple/dev-project-coordinator ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model21 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubDev Project Coordinator
Local desk for agents. Track coding and non coding work as JSON on your machine. Agents keep the board. You open a local Analytics tab and Kanban tab.
Formerly Claude Project Coordinator.
Give your AI a project board it reads at the start of every chat, so work survives between sessions and between agents.
What this does
Your AI agents keep a small Kanban of every project you run, coding or not: what is in progress, what is blocked, what is waiting for review, and the commit or link that proves something shipped. When a new chat starts, the agent asks the desk for a summary of under 800 tokens and gets back on track, instead of guessing or rereading your repo. You do not need a memory or RAG tool for this. It works on its own, provided the agent's rules tell it to read the desk at the start of each session. If you do use a memory tool, the two sit side by side: memory keeps the why, DPC keeps the what and where. Everything stays as JSON on your own disk, and you get a local Analytics and Kanban page where you can drag cards yourself.
What you can ask
Type these as you would to any assistant. Each one maps to a real DPC tool.
| You type | Tool | What happens |
|---|---|---|
| "What is on my desk?" | get_desk_summary |
The compact state of every project. Ask this first in each session so the agent starts from the truth, not from memory |
| "I am in this repo. Register it." | register_from_workspace |
Reads the folder and git remote, detects the language, and adds a starter card. It never creates folders |
| "Add a card for the failing tests and put it in doing." | add_card, move_card |
A new card in the doing column of the current project |
| "Move the login fix to review, note commit 299b827." | move_card with note |
The card moves and carries a short evidence line, so the next session or agent sees what was done |
| "File this under a legal project called Q4 review." | file_under_project |
Finds or creates a project with no folder, for writing, legal, accounting or other non coding work |
| "Which projects need attention?" | get_project_health, get_activity_heatmap |
Health scores with reasons, and where activity has gone quiet |
How it works
Everyday flow. You talk to your AI, the AI reads and updates the desk, and the desk is plain JSON you can also open as a page.
flowchart LR
U[You] --> A[Your AI<br/>Claude, Cursor or Grok]
A -->|reads the desk first| D[DPC]
A -->|adds and moves cards| D
D --> J[(JSON on your disk)]
J --> K[Kanban and Analytics page]
Beside a memory or RAG tool. DPC holds the live board, memory holds the reasons. The only link is a memoryRef id on a card. DPC never calls the memory tool. With no memory tool, leave memoryRef empty and DPC works alone.
flowchart LR
A[Your AI] -->|what and where<br/>cards, columns, counts| D[DPC]
A -->|why<br/>decisions, context| M[Memory or RAG tool]
D -.->|memoryRef id only| M
Handoff between sessions or agents. One session or agent leaves a card in review with a note, the next one reads the desk summary and carries on. To share one desk across machines, sync its folder with git (see the shared board guide).
flowchart LR
S1[Session 1 or Agent A] -->|card to review<br/>with commit note| D[(Shared desk)]
D -->|desk summary| S2[Session 2 or Agent B]
S2 -->|card to done| D
Who this is for
People who run AI agents across several projects and are tired of each new chat starting cold. People who mix code with writing, legal, accounting or campaign work in the same chat. Anyone running several agents who needs one truthful board they all read and update. And anyone who wants this kept local, with no account and no telemetry.
How it fits with memory and RAG
| DPC | Memory / RAG | |
|---|---|---|
| Job | Live structured state: column, counts, board | Durable why, decisions, people |
| Shape | Exact rows and a Kanban | Similar chunks |
| Ask | "What is blocked right now?" | "Why did we choose X last month?" |
Put only a memoryRef (opaque id or tag) on a card. DPC never calls a memory service and never copies the prose. If you use RAG only, exact counts and a current board are not available. If you use neither, DPC still keeps your agents on track between sessions.
What it is not
- Not a memory or RAG product. There are no embeddings and no semantic search.
search_deskis a plain text match over notes, cards, kinds and markdown.- No live sync. Sharing a board uses git push and pull. Merges are a union by card id.
- No Trello, Asana or Jira sync. Use those vendors' own MCPs beside DPC.
memoryRefis only an id. DPC cannot fetch the decision it points to.- Nothing reads the desk automatically. Your host's rules or prompt (for example
CLAUDE.md,AGENTS.mdor docs/grok-bot/PROMPT.md) must tell the agent to callget_desk_summaryat the start. - The desk server is local only, bound to
127.0.0.1. Cloud agents reach a desk through the shared board route, not the Dock app.
DPC is a local stdio MCP server. It stores live project state (column, kind, counts, cards) as JSON. It does not host your data. It does not replace a memory or RAG product.
v2.0.2 · official Swift MCP SDK · Apache 2.0 · local JSON knowledge base
Quick start
Needs macOS 13+, Swift 6.0+ (Xcode 16+) and an MCP host (Claude Desktop, Cursor, Claude Code, or Grok Build). Linux runs through the Dockerfile.
-
Build:
git clone https://github.com/M-Pineapple/Dev-Project-Coordinator.git cd Dev-Project-Coordinator swift build -c release -
Add the server. The binary name stays
project-coordinator.dpcis an alias.{ "mcpServers": { "project-coordinator": { "command": "/absolute/path/to/Dev-Project-Coordinator/.build/release/project-coordinator", "args": [] } } } -
Restart the host. Ask: "What is on my desk?" The agent should call
get_desk_summaryfirst.
Coding folders must sit under an allowed root. Non coding work can have no path.
Open it as a Mac app
The desk page is a local file by default. For drag-and-drop, and to pin it in the Dock like an app:
- Ask the agent to start the desk server, or run
dpc start_desk_server. - Open the sign-in URL once in Safari. It carries a one-time code, valid once, for 10 minutes. That sets a session cookie (the write token) and redirects to
http://127.0.0.1:47821/with no code in the address bar. - Safari: File > Add to Dock. The Dock shortcut is the plain URL. It does not store the code or the write token.
The port is fixed at 47821. Change it with DPC_DESK_PORT or deskPort in security-config.json. If that port is busy, start fails with a clear error. It does not pick a random port.
The write token persists in KnowledgeBase/desk/.write-token, so restarting the desk server keeps the Dock app signed in. On start the server reuses that file only when it is a regular file owned by you, mode 0600, holding a 64 character hex token. Otherwise it mints a new token, replaces the file and logs a warning, and you open a new sign-in URL once.
Call start_desk_server again while it is running to mint a fresh sign-in code without rotating the write token. Rotate the write token with start_desk_server rotate=true, or dpc start_desk_server --rotate, on a running or a fresh server. Open the new sign-in URL once so the Dock app cookie updates.
Starting points by setup
1. Agent user with several repos
You already have coding agents. You do not want a second project-management SaaS.
Start with:
- "I am in this repo. Register it."
- "What is blocked right now?"
- "Add a card for the failing tests and put it in doing."
What you get: silent register from cwd and git remote, a compact desk under 800 tokens, cards linked to a PR when you pass one.
2. Mixed coding and non coding desk
You also run accounting, legal, writing, or campaign work in the same chat.
Start with:
- "File this under a legal project called Q4 review."
- "Rename that to Autumn review."
- "Merge Duplicate Campaign into Autumn review."
What you get: find-or-create with no folder, a one-line notice Filed under new project: …, and rename/merge when the guess is wrong.
3. Privacy first local setup
You will not send a board to a hosted PM tool.
Start with:
- Build from source. Point the host at
.build/release/project-coordinator. - Optional:
DPC_KNOWLEDGE_BASEfor data outside the git tree.CPC_KNOWLEDGE_BASEstill works through v2.x. - Open
KnowledgeBase/desk/index.htmlfrom disk.
What you get: JSON on your disk, path allowlisting, no account, no telemetry.
4. Already using a memory or RAG product
You store decisions in a memory MCP. You do not want that text copied here.
How they fit together: see How it fits with memory and RAG. Put only a memoryRef on a card. DPC never copies the prose.
5. Grok Bot user
You want a specialised desk operator.
Start with: paste docs/grok-bot/PROMPT.md, then add this MCP by hand. Templates cannot carry a custom MCP.
What you get: the same JSON desk. An optional private Grok Bot Page is a view, not a second store. Never auto-share.
Full feature list
Desk
| Capability | Detail |
|---|---|
| Compact summary | get_desk_summary, hard cap 800 tokens (~3,200 characters) |
| Any language | Shallow detector: swift, python, javascript, typescript, node, go, rust, ruby, php, java, kotlin, csharp, dotnet, cpp, docker, web, plus spm / xcode-project / xcode-workspace / swiftui from top-level files |
| Non coding kinds | writing, campaign, research, legal, accounting, business, personal, other. Path optional |
| Auto-register | From cwd + git remote + optional open PR. Never creates folders. Adds a starter card |
| Find or create | Normalised name match, near-duplicate guard, rename and merge. Adds a starter card |
| Kanban | Columns backlog, doing, review, done, parked. Users may rename |
| Cards | id, project, title, column, order, due, optional memoryRef, optional note. note is omitted when empty. No archive field is stored |
| Local page | Analytics tab + Kanban tab as KnowledgeBase/desk/index.html |
| Interactive board | Opt-in start_desk_server on 127.0.0.1:47821, 0600 write token, Safari Add to Dock |
Done cards whose updated time is older than 14 days are hidden on the Done column. Show archived (N) reveals them, muted. The Done in 7 days stat and the analytics tab still count every card. Nothing is deleted, and cards.json does not gain an archive field, so a shared-board merge stays a union by card id. Set the window with DPC_ARCHIVE_DAYS or archiveDays in security-config.json. get_desk_summary and get_project_status skip archived cards and still report the count. Pass includeArchived to list them. search_desk still finds them and marks them archived. Moving a card out of done, or any edit that refreshes updated, brings it back.
A card may also carry an optional note, a short evidence line such as Tech: commit 299b827 or a URL. Set it with add_card, update_card, or move_card. The board shows it in small muted text, and search_desk matches it. Leaving note off a move keeps the current line. A blank note clears it.
Analytics tab
The same JSON the MCP tools read, rendered as a local dashboard (file and live modes):
| Section | What you see |
|---|---|
| Attention | Five items that need a look, ranked |
| Board flow | Column mix, stale cards (14 days), 7-day throughput into done, parked count |
| Kind mix | Coding vs writing, legal, accounting, and the rest |
| Project health | 0 to 100 score per project, factors, recommendations, grouped critical / needs attention / healthy. Non coding projects are not scored down for missing code docs or a tech stack |
| Activity heat map | Per project, last 14 days, CSS intensity cells and a sparkline. Total events. No emoji |
| Technology trends | Share bars, emerging tags (1 or 2 coding projects), adoption timeline. Coding projects only |
| Project timeline | Created date, current status, time in each status. Collapsible per project |
No emoji as data. No remote fonts or assets. Light and dark follow the system.
Safety and storage
| Capability | Detail |
|---|---|
| Local JSON | KnowledgeBase/projects/ and KnowledgeBase/desk/ |
| Path allowlist | Compiled roots plus SecurityConfig.allowedPaths and DPC_ALLOWED_PATHS |
| Input limits | Name, path, notes, and search length capped |
| HTML | User strings escaped on the desk page |
| Env | DPC_KNOWLEDGE_BASE preferred, CPC_KNOWLEDGE_BASE until v3.0 |
| Stdio-safe logs | Diagnostics go to stderr only, prefix [DPC] |
All MCP tools
| Tool | Parameters | Purpose |
|---|---|---|
get_desk_summary |
optional includeArchived |
Compact desk. Use this first. Archived done cards are omitted unless includeArchived is true. The archived count is always reported |
list_projects |
none | Full roster. Deep dive only |
get_project_status |
projectName, optional includeArchived |
Full detail for one project. The card list skips archived done cards unless includeArchived is true |
add_project |
name, optional path, description, kind |
Manual add |
update_project_status |
projectName, optional status, notes |
Update the desk entry |
search_desk |
pattern |
Search notes, cards, kinds, markdown. Archived cards are included and marked |
search_code_patterns |
pattern |
Deprecated alias of search_desk |
get_project_timeline |
projectName |
Status history and time in each status |
get_activity_heatmap |
optional days (default 7) |
Activity heat map (points, events, daily counts) |
get_technology_trends |
none | Framework share, emerging tags, adoption timeline (coding only) |
get_project_health |
none | 0 to 100 scores, factors, recommendations |
open_desk |
none | Write desk/index.html and return its path |
add_card |
project, title, optional column, memoryRef, linkedPR, note |
Add a card. note is a short evidence line |
move_card |
id, column, optional note |
Move a card. Omit note to keep the current one |
update_card |
id, optional title, column, memoryRef, note |
Edit a card. A blank note clears it |
rename_column |
from, to |
Rename a board column |
register_from_workspace |
cwd, optional gitRemote, branch, openPR |
Silent coding register |
file_under_project |
name, optional kind, description |
Non coding find or create |
rename_project |
from, to |
Fix a wrong name |
merge_projects |
source, target |
Absorb a duplicate |
start_desk_server |
optional rotate |
Opt-in loopback Kanban. Returns a bookmark URL and a one-time sign-in URL (valid once, for 10 minutes). Call again to mint a new code without rotating |
You talk in natural language. The host maps your words onto these tools.
Example workflow
You: I am in ~/Developer/widget-kit on branch main, PR 42. Register it.
Agent: calls register_from_workspace. Detected javascript.
You: What needs me?
Agent: calls get_desk_summary. Compact table, no notes dump.
You: File the invoice chase under a new accounting project.
Agent: calls file_under_project. "Filed under new project: Invoice chase"
You: Open the desk.
Agent: calls open_desk. You open KnowledgeBase/desk/index.html from disk.
Prerequisites
- macOS 13+ for the default path
- Swift 6.0+ (Xcode 16+)
- An MCP host
- Optional: Docker for the Linux image
Installation
Build from source
git clone https://github.com/M-Pineapple/Dev-Project-Coordinator.git
cd Dev-Project-Coordinator
swift build -c release
Executables:
.build/release/project-coordinator
.build/release/dpc
Claude Desktop
Settings → Developer → Model Context Protocol. Use the JSON in Quick start. No --port flag. This is stdio, not HTTP.
Cursor
Add the same server entry to your Cursor MCP settings JSON.
Claude Code
Register the same command path.
Grok Build
grok mcp add dpc -- /absolute/path/to/Dev-Project-Coordinator/.build/release/project-coordinator
Grok Bot
Create a Dev Project Coordinator bot in your Grok Bot:
- Create a new bot, or ask the bot you use to create bots to make one, and give it the prompt below as its instructions.
- Add this server to that bot as a custom MCP (see the install notes in the prompt). Bot templates cannot carry a custom MCP, so this step is always manual.
- Tell the bot what you are working on. It registers projects for you, keeps the board current, and can show the Analytics and Kanban views as a private Page.
## Identity
You are a desk operator for Dev Project Coordinator (DPC). You keep a local JSON board truthful. You do not write code in the user's repos. You do not host user data. You do not replace a memory or RAG product.
## Session start
1. Call `get_desk_summary` first.
2. Do not call `list_projects` unless the user asks for a deep dive.
3. If this is a coding session and you have a working directory, call `register_from_workspace` with cwd, git remote, current branch, and the open PR if you have it. Never create folders on disk.
4. If this is non-coding work, call `file_under_project`. If the tool says `Filed under new project: <name>`, tell the user that line. If the guess is wrong, call `rename_project` or `merge_projects` once.
## Board rules
- One card, one owner, one column.
- Columns default to backlog, doing, review, done, parked. Rename only when the user asks.
- Done cards older than the archive window (14 days unless configured) are hidden on the board and left out of `get_desk_summary` and `get_project_status`. They are not deleted. `search_desk` still finds them. Move one out of done to bring it back.
- A card may have an optional `note`: a short evidence line (a commit or a URL), not a memory dump. Set it with `add_card`, `update_card`, or `move_card`.
- After any card or status write, call `open_desk` so `KnowledgeBase/desk/index.html` is current.
- Ask before deleting or merging.
## Memory
If the user has a memory or RAG product, write the decision there. Set `memoryRef` on the DPC card. Put no prose from memory into DPC. If they have no memory product, leave `memoryRef` empty. DPC still works.
## Grok Bot Pages
You may refresh a **private** Page from `get_desk_summary` as an Analytics view and a Kanban view. The Page is a view, not a store. Do not auto-share. Project names and card titles are private. A public `pages.grok.bot` link is a publish action and needs an explicit user sentence that names that action. Local `index.html` remains the view for every other host.
## MCP install (tell the user)
Templates do not carry this server. After import, add a custom MCP named `dpc` (or `project-coordinator`) that runs the local binary:
```text
/absolute/path/to/Dev-Project-Coordinator/.build/release/project-coordinator
```
On Grok Bot's cloud computer, sibling bots can read files. Prefer local-computer execution, or the Linux Docker image with `DPC_KNOWLEDGE_BASE` on that disk.
## Shared board (optional)
If the user wants one board shared between your computer and their desktop, or readable by another assistant, offer the optional shared-board setup: a private GitHub repo as transport, a locked wrapper on your side, a two-way mirror on their desktop, and an optional read-only viewer. Follow docs/shared-board/README.md in the DPC repo step by step and do not improvise your own sync. Never put a token or key in a card, note, or message.
## Anti-jobs
- Do not implement Trello, Asana, or Jira.
- Do not copy memory prose into cards or notes.
- Do not start `start_desk_server` unless the user asks for drag-and-drop.
- Do not claim the template "just works" without the MCP.
The matching skill is in docs/grok-bot/SKILL.md.
Templates cannot include this server. After you paste docs/grok-bot/PROMPT.md, tell the bot to add a custom MCP that runs the binary. On the bot cloud computer, sibling bots can read files. Prefer local execution, or the Docker image with DPC_KNOWLEDGE_BASE on that disk.
Optional: set DPC_KNOWLEDGE_BASE (or legacy CPC_KNOWLEDGE_BASE) to a custom KnowledgeBase directory.
Optional: to share one board between a Linux host (such as a Grok Bot's computer) and your desktop, see the shared board guide.
Restart the host after changing the config.
Dev Project Coordinator MCP questions
What tools does Dev Project Coordinator MCP provide?
It exposes 21 tools, including get_desk_summary, list_projects, get_project_status, add_project, update_project_status.
What can I ask my AI to do with Dev Project Coordinator?
You can ask it to register a repo, add or move cards, file work under non-coding projects like legal or accounting, and report which projects need attention. Each request maps to a real tool such as add_card, move_card, register_from_workspace or get_project_health.
Where is my project data stored?
All project state is kept as JSON on your own disk with no account and no telemetry. You can also open it as a local Analytics and Kanban page.
Is this a memory or RAG tool?
No. It holds live structured board state, not embeddings or semantic search, and questions like why a decision was made belong in a separate memory tool. A card can carry only a memoryRef id pointing at that other tool.
Does it sync with Trello, Asana or Jira?
No. There is no sync with those vendors, and sharing a board across machines works by syncing its folder with git.
What does it need to run?
It requires macOS 13+, Swift 6.0+ and an MCP host such as Claude Desktop, Cursor, Claude Code or Grok Build. Linux runs through the Dockerfile.