@selfagency/beans-mcp 🫘

<img src="docs/assets/icon.png" alt="beans-mcp icon" width="300" />

MCP (Model Context Protocol) server for Beans issue tracker. Provides programmatic and CLI interfaces for AI-powered interactions with Beans workspaces.

Documentation: beans-mcp.self.agency

🤖 Try Beans fully-integrated with GitHub Copilot in VS Code! Install the <a href="https://marketplace.visualstudio.com/items?itemName=selfagency.beans-vscode">selfagency.beans-vscode</a> extension.

Usage

npx @selfagency/beans-mcp /path/to/workspace

Versioning

@selfagency/beans-mcp has its own package versioning. Compatibility with the Beans CLI is tracked separately.

At startup, the server compares the installed beans CLI version against the hardcoded supported Beans version: 0.4.2. If they differ, it prints a warning to stderr and continues startup.

Parameters

• --workspace-root or positional arg: Workspace root path
• --cli-path: Path to Beans CLI
• --port: MCP server port (default: 39173)
• --log-dir: Log directory
• -h, --help: Print usage and exit

Summary of public MCP tools

Tool	Description
beans_init	Initialize the workspace (optional prefix).
beans_archive	Archive completed/scrapped beans.
beans_view	Fetch full bean details by beanId or beanIds.
beans_create	Create a new bean (title/type + optional body/parent).
beans_bulk_create	Create multiple beans in one call, optionally under a shared parent.
beans_update	Consolidated metadata + body updates (status/type/priority/parent/clearParent/blocking/blockedBy/body/bodyAppend/bodyReplace) plus optional optimistic concurrency hint (ifMatch).
beans_bulk_update	Update multiple beans in one call, optionally reassigning them to a shared parent.
beans_complete_tasks	Mark all markdown checklist tasks within a bean as complete.
beans_delete	Delete one or many beans (beanId or beanIds, optional force).
beans_reopen	Reopen a completed or scrapped bean to an active status.
beans_query	Unified list/search/filter/sort/ready operations, with GraphQL passthrough.
beans_bean_file	Read/edit/create/delete files under .beans.
beans_output	Read extension output logs or show guidance.

<details> <summary>Notes</summary>

• The beans_query tool is intentionally broad: prefer it for listing, searching, filtering or sorting beans, and for generating Copilot instructions (operation: 'llm_context').
• All file and log operations validate paths to keep them within the workspace or the VS Code log directory. The .beans/ prefix is automatically stripped from paths — you can pass either some-bean.md or .beans/some-bean.md and the result is the same.
• beans_update replaces many fine-grained update tools; callers should use it to keep the public tool surface small and predictable.
• beans_archive provides CLI parity for archiving completed/scrapped beans.
• Closing a parent bean via beans_update (status: completed or status: scrapped) cascades the same status to all descendants.
• Reopening a parent bean via beans_reopen cascades the target status to closed descendants (completed / scrapped).
• beans_bulk_create and beans_bulk_update are best-effort: they process each item sequentially and return a per-item result array with success/error entries rather than failing atomically.
• Frontmatter title: values are automatically double-quoted on write. Pass raw titles — quoting and escaping is handled for you.
• beans_bean_file supports update_frontmatter for atomic frontmatter-only writes; supported fields include pr and branch.
• Unfiltered list results are cached with a short burst TTL and a timestamp-probe refresh strategy. Mutation tools (beans_create, beans_update, beans_delete, etc.) invalidate the cache immediately.
• Version mismatches between beans-mcp and the Beans CLI are warning-only and non-blocking by design.
• When beanId is missing in tool input, validation errors include a hint: Did you mean \beanId`?`.

</details>

Examples

<details> <summary>beans_init</summary>

Request:

{ "prefix": "project" }

Response (structuredContent):

{ "initialized": true }

</details>

<details> <summary>beans_view</summary>

Request:

{ "beanId": "bean-abc" }

Request (multiple beans):

{ "beanIds": ["bean-abc", "bean-def"] }

Response (structuredContent):

{
  "bean": {
    "id": "bean-abc",
    "title": "Fix login timeout",
    "status": "todo",
    "type": "bug",
    "priority": "critical",
    "body": "...markdown...",
    "createdAt": "2025-12-01T12:00:00Z",
    "updatedAt": "2025-12-02T08:00:00Z"
  }
}

</details>

<details> <summary>beans_archive</summary>

Request:

{}

Response (example):

{ "archived": true, "archivedCount": 3 }

</details>

<details> <summary>beans_create</summary>

Request:

{
  "title": "Add dark mode",
  "type": "feature",
  "status": "todo",
  "priority": "normal",
  "body": "Implement theme toggle and styles",
  "parent": "epic-123"
}

description is accepted as a deprecated alias for body.

Response (structuredContent):

{
  "bean": {
    "id": "new-1",
    "title": "Add dark mode",
    "status": "todo",
    "type": "feature"
  }
}

</details>

<details> <summary>beans_bulk_create</summary>

Request:

{
  "parent": "epic-123",
  "beans": [
    { "title": "Design mockups", "type": "task" },
    { "title": "Implement API", "type": "task", "priority": "high" },
    { "title": "Write tests", "type": "task", "parent": "epic-456" }
  ]
}

The top-level parent is applied as a default to any bean that does not specify its own parent. Here Design mockups and Implement API are assigned to epic-123; Write tests overrides with epic-456.

Response (structuredContent):

{
  "requestedCount": 3,
  "successCount": 3,
  "failedCount": 0,
  "results": [
    { "bean": { "id": "task-1", "title": "Design mockups" } },
    { "bean": { "id": "task-2", "title": "Implement API" } },
    { "bean": { "id": "task-3", "title": "Write tests" } }
  ]
}

</details>

<details> <summary>beans_bulk_update</summary>

Request (move a batch of tasks to in-progress and assign them to a parent):

{
  "parent": "epic-123",
  "beans": [
    { "beanId": "task-1", "status": "in-progress" },
    { "beanId": "task-2", "status": "in-progress" },
    { "beanId": "task-3", "status": "in-progress", "parent": "epic-456" }
  ]
}

Response (structuredContent):

{
  "requestedCount": 3,
  "successCount": 3,
  "failedCount": 0,
  "results": [
    { "beanId": "task-1", "bean": { "id": "task-1", "status": "in-progress" } },
    { "beanId": "task-2", "bean": { "id": "task-2", "status": "in-progress" } },
    { "beanId": "task-3", "bean": { "id": "task-3", "status": "in-progress" } }
  ]
}

Both bulk tools are best-effort: partial failures are reported per-item rather than aborting the whole batch.

</details>

<details> <summary>beans_update</summary>

Request (change status and add blocking):

{
  "beanId": "bean-abc",
  "status": "in-progress",
  "blocking": ["bean-def"],
  "ifMatch": "etag-value"
}

Request (atomic body modifications):

{
  "beanId": "bean-abc",
  "bodyReplace": [
    { "old": "- [ ] Task 1", "new": "- [x] Task 1" },
    { "old": "- [ ] Task 2", "new": "- [x] Task 2" }
  ],
  "bodyAppend": "## Summary\n\nAll checklist items completed."
}

Note: body (full replacement) cannot be combined with bodyAppend or bodyReplace in the same request.

Response (structuredContent):

{
  "bean": {
    "id": "bean-abc",
    "status": "in-progress",
    "blockingIds": ["bean-def"]
  }
}

</details>

<details> <summary>beans_delete</summary>

Request:

{ "beanId": "bean-old", "force": false }

Response:

{ "deleted": true, "beanId": "bean-old" }

Batch request:

{ "beanIds": ["bean-old", "bean-older"], "force": false }

Batch response (summary):

{
  "requestedCount": 2,
  "deletedCount": 2,
  "failedCount": 0,
  "results": [
    { "beanId": "bean-old", "deleted": true },
    { "beanId": "bean-older", "deleted": true }
  ]
}

</details>

<details> <summary>beans_reopen</summary>

Request:

{
  "beanId": "bean-closed",
  "requiredCurrentStatus": "completed",
  "targetStatus": "todo"
}

Response:

{ "bean": { "id": "bean-closed", "status": "todo" } }

</details>

<details> <summary>beans_complete_tasks</summary>

Request:

{ "beanId": "bean-abc" }

Response:

{
  "bean": {
    "id": "bean-abc",
    "status": "todo"
  },
  "totalTaskCount": 5,
  "updatedTaskCount": 3,
  "unchangedTaskCount": 2
}

</details>

<details> <summary>beans_query examples</summary>

Refresh (list all beans):

{ "operation": "refresh" }

Response (partial):

{ "count": 12, "beans": [] }

Filter (statuses/types/tags):

{
  "operation": "filter",
  "statuses": ["in-progress", "todo"],
  "types": ["bug", "feature"],
  "tags": ["auth"]
}

Search (full-text):

{ "operation": "search", "search": "authentication", "includeClosed": false }

Sort (modes: status-priority-type-title, updated, created, id):

{ "operation": "sort", "mode": "updated" }

Ready (actionable beans only):

{ "operation": "ready" }

LLM context (generate Copilot instructions; optional write-to-workspace):

{ "operation": "llm_context", "writeToWorkspaceInstructions": true }

Response (structuredContent):

{
  "graphqlSchema": "...",
  "generatedInstructions": "...",
  "instructionsPath": "/workspace/.github/instructions/beans-prime.instructions.md"
}

Raw GraphQL passthrough (CLI parity with beans query):

{
  "operation": "graphql",
  "graphql": "{ beans(filter: { type: [\"bug\"] }) { id title status } }"
}

With variables:

{
  "operation": "graphql",
  "graphql": "query($q: String!) { beans(filter: { search: $q }) { id title } }",
  "variables": { "q": "authentication" }
}

</details>

<details> <summary>beans_bean_file</summary>

Request (read):

{ "operation": "read", "path": "beans-vscode-123--title.md" }

Response:

{
  "path": "/workspace/.beans/beans-vscode-123--title.md",
  "content": "---\n...frontmatter...\n---\n# Title\n"
}

Request (atomic frontmatter update):

{
  "operation": "update_frontmatter",
  "path": "beans-vscode-123--title.md",
  "fields": {
    "status": "in-progress",
    "pr": "123",
    "branch": "feature/cascade-status-and-skills-npm"
  }
}

Response:

{
  "path": "/workspace/.beans/beans-vscode-123--title.md",
  "bytes": 256,
  "updatedFields": ["status", "pr", "branch"],
  "frontmatter": {
    "status": "in-progress",
    "pr": "123",
    "branch": "feature/cascade-status-and-skills-npm"
  }
}

</details>

<details> <summary>beans_output</summary>

Request (read last 200 lines):

{ "operation": "read", "lines": 200 }

Response:

{
  "path": "/workspace/.vscode/logs/beans-output.log",
  "content": "...log lines...",
  "linesReturned": 200
}

</details>

Programmatic usage

Installation

npm install beans-mcp

Example

import { createBeansMcpServer, parseCliArgs } from '@selfagency/beans-mcp';

const server = await createBeansMcpServer({
  workspaceRoot: '/path/to/workspace',
  cliPath: 'beans', // or path to beans CLI
});

// Connect to stdio transport or your own transport

API

createBeansMcpServer(opts)

Creates and initializes a Beans MCP server instance.

Options:

• workspaceRoot (string): Path to the Beans workspace
• cliPath (string, optional): Path to Beans CLI executable (default: 'beans')
• name (string, optional): Server name (default: 'beans-mcp-server')
• version (string, optional): Server version
• logDir (string, optional): Directory for server logs
• backend (BackendInterface, optional): Custom backend implementation

Returns: { server: McpServer; backend: BackendInterface }

startBeansMcpServer(argv)

CLI-compatible entrypoint for launching the server.

Utility Functions

• parseCliArgs(argv: string[]): Parse CLI arguments
• isPathWithinRoot(root: string, target: string): boolean: Check if path is contained within root
• sortBeans(beans, mode): Sort beans by specified mode

Types & Schemas

Export of GraphQL schema, Zod validation schemas, and TypeScript types for Beans records and operations.

Agent Skills (skills-npm, skills.sh)

This package ships a built-in Agent Skill under skills/ and also publishes that skill in a format that fits the broader open skills ecosystem surfaced by skills.sh.

• Skill path in package: skills/beans-mcp/SKILL.md
• Published skill artifact: https://beans-mcp.self.agency/.well-known/agent-skills/beans-mcp/SKILL.md
• Published discovery index: https://beans-mcp.self.agency/.well-known/agent-skills/index.json
• Compatible with discovery tools that scan: node_modules/**/skills/*/SKILL.md

That means you can use it with npm-based workflows such as skills-npm, while also pointing ecosystem tooling at the published skill artifact and discovery index used by skills catalogs like skills.sh.

To symlink installed npm-packaged skills into your agent workspace, you can use skills-npm in your consuming project.

License

MIT