Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@agent-gantt Create a project 'Website Redesign' with tasks: design, develop, test."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agent-gantt is a dependency-free Model Context Protocol server for maintaining small project plans that AI agents can render as Gantt charts. State lives in SQLite and the server communicates over MCP stdio.
Projects, tasks, assignees, status, progress, and date ranges.
Explicit same-project task dependencies; nonexistent tasks and dependency cycles are rejected.
Gantt-friendly project view with tasks and dependency edges.
Python 3.9+ and its standard library only. No package installation or network service is required.
Related MCP server: Project Tracking MCP Server
Requires Python 3.9+ and uv. Register the package directly from GitHub with the MCP client’s normal stdio configuration. The first launch installs the package in an isolated environment; no repository checkout or custom installer is needed.
# Claude Code
claude mcp add -s user agent-gantt -- uvx --from git+https://github.com/TobiasCoding/agent-gantt-mcp.git agent-gantt
# Codex CLI
codex mcp add agent-gantt -- uvx --from git+https://github.com/TobiasCoding/agent-gantt-mcp.git agent-ganttThe default database is ~/.local/state/agent-gantt/projects.sqlite3. To share a database across clients, add AGENT_GANTT_DB to that server’s MCP env configuration (or use the client CLI’s environment option).
create_project, list_projects
create_task, update_task
add_dependency
project_view returns all tasks and dependency edges in a Gantt-friendly structure.
Dates are ISO dates (YYYY-MM-DD). Valid status values are todo, in_progress, blocked, and done.
The repository contains no project, task, or timeline data. At runtime, data is written to ~/.local/state/agent-gantt/projects.sqlite3 by default; it is intentionally ignored by Git. Treat that database as sensitive if it contains planning information. Set AGENT_GANTT_DB to choose another durable location.
MIT. See LICENSE.
Add predecessor; rejects cycles and cross-project edges.
Inputs: task_id (required), depends_on_id (required), actor, reason, +1 more.
Atomic 1..50 task/dependency operations. Each has tool, arguments, optional ref on create_task. Use $ref in later task_id/depends_on_id.
Inputs: operations (required), idempotency_key.
Read unfiltered task changes after_revision. Apply by task ID; after all pages poll from through_revision. Get full task using event revision.
Inputs: project_id (required), limit, cursor, after_revision.
Atomic task lifecycle with fresh revision. Claim checks assignment/dependencies. Owner can block/resume/report progress/release/complete; reason records evidence.
Inputs: actor (required), action (required), reason (required), task_id (required), +3 more.
Create project.
Inputs: name (required), description, idempotency_key.
Create task. summary: few words for humans; description: agent detail. parent_task_id groups subtasks, not a dependency.
Inputs: title (required), project_id (required), actor, reason, +9 more.
Retrieve complete JSON in segments; concatenate json_segment using next_offset. For stable task reads use kind=event and revision as id.
Inputs: id (required), kind (required), count, offset.
List project IDs/names; detail on demand.
Inputs: limit, cursor.
Paginated snapshot. Save revision. ready_only finds executable tasks; assignee includes unassigned when ready_only.
Inputs: project_id (required), limit, cursor, status, +2 more.
Update supplied fields; null clears a date. Claimed tasks require actor=owner.
Inputs: task_id (required), actor, title, reason, +9 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 10 tools
There is real overlap between mutation tools (update_task vs coordinate, which both change task state/lifecycle) and among read tools (list_projects, project_view, changes, get_detail all surface state). Descriptions help distinguish them (snapshot vs. delta vs. raw JSON), but an agent must read carefully to choose the right read/write path.
Mostly verb_noun (create_project, create_task, update_task, add_dependency, list_projects), but the set mixes in bare verbs (coordinate, changes, batch) and a noun_verb (project_view). Readable, but not a predictable single pattern.
10 tools is well-scoped for a Gantt/project-management server, covering project, task, dependency, lifecycle, batch, and read/sync operations without bloat.
Create/update exist for tasks and create/list for projects, but there is no delete_task, delete/update_project, or remove_dependency (add_dependency has no inverse). Read/sync coverage is strong, but lifecycle gaps could force agents into workarounds.
Project management MCP for AI agents with safe task reads and writes.
Project management shared by people and AI agents, with persistent project state through MCP.
Minimal issue tracker: projects, objectives and trackable plans, written by agents over MCP.
Project tracker for AI agents and teams: file, claim, update and hand off tickets over MCP.
