Capacities MCP
An MCP server for Capacities API 2.0 that enables AI agents to discover, create, read, update, and delete objects, blocks, and daily notes with structured JSON or Markdown, using personal API tokens.
README
Capacities MCP Enhanced — API 2.0 for AI Agents
The official MCP server sucks, it's time to make a new one.
Built for the new Capacities API 2.0 with native precise block level CRUD support.
Capacities MCP brings rich, read-write access to your knowledge base through one agent-friendly server. Agents can discover your object types, understand custom properties, create polished content, make precise edits, save web resources, and work with Daily Notes.
✨ Why Capacities MCP Enhanced
- Two first-class authoring engines — fast Markdown writing and precise structured blocks in one server.
- Full CRUD with Capacities API 2.0 — objects, custom properties, collections, blocks, web links, tasks, tags, and Daily Notes.
- Designed for autonomous agents — live schema discovery, consistent responses, focused tools, and clear recovery information.
- Safer writes — independent readback verification, soft deletion by default, API-key rate-limit failover, and rollback for failed multi-stage Markdown creates.
- Concurrency-safe object operations — same-object reads can run concurrently, while writes use an exclusive lock and automatically wait for active reads or other writes; different objects proceed concurrently.
- Faithful rich content — headings, nested blocks, grids, toggles, tasks, links, entity references, colors, code, and math.
- Agent-friendly media uploads — stream one or many local files, report progress, verify media objects, and manage long-running jobs without exposing multipart internals.
- Bundled best-practice skill — a complete agent operating guide is included, so a model can use the server correctly without already knowing Capacities.
🚀 Quick start
Requirements
- Node.js 20+ or Bun 1.3+
- One or more personal API keys from the new Capacities API with the permissions you need
Installation
Configure your MCP client to use the Capacities MCP server:
Generate a token in Capacities → Settings → Capacities API → Generate new token. New API tokens begin with cap-api-.
Configuring at least 3 tokens with same space and permissions is recommended to avoid rate-limit errors.
Claude Code
{
"mcpServers": {
"capacities": {
"command": "npx",
"args": ["--yes", "capacities-mcp-enhanced@latest"],
"env": {
"CAPACITIES_API_TOKEN": "cap-api-key-1,cap-api-key-2;cap-api-key-3"
}
}
}
}
Hermes Agent
mcp_servers:
capacities:
command: "npx"
args:
- "--yes"
- "capacities-mcp-enhanced@latest"
env:
CAPACITIES_API_TOKEN: "cap-api-key-1,cap-api-key-2;cap-api-key-3"
supports_parallel_tool_calls: true
Install the bundled skills (Optional)
By installing the Capacities MCP Best Practices skill, your agent can learn the full workflow, strict content schemas, safe mutation patterns, rate-limit recovery, and common pitfalls.
npx skills add nyrest/capacities-mcp-enhanced
🧰 What agents can do
| Category | Tool | What it does |
|---|---|---|
| Discover and read | inspect_space |
Discover structures, properties, labels, and available write options. |
| Discover and read | search_objects |
Find objects by title, optionally within selected structures. |
| Discover and read | get_object |
Read a complete object as structured content or Markdown. |
| Create objects | create_object |
Create a Page, Task, Tag, or custom object with structured content. |
| Create objects | create_object_markdown |
Create an object from a Markdown document. |
| Create objects | create_object_from_url |
Save a web resource with structured notes. |
| Create objects | create_object_from_url_markdown |
Save a web resource with Markdown notes. |
| Update content | update_object |
Update titles, typed properties, and collections. |
| Update content | append_content |
Add structured blocks to an object. |
| Update content | append_content_markdown |
Add Markdown content to an object. |
| Update content | update_block |
Precisely edit one existing block. |
| Update content | delete_block |
Remove one block and its nested content. |
| Update content | delete_object |
Move an object to trash or permanently delete it when explicitly requested. |
| Daily Notes | append_daily_note |
Add structured blocks to today's or a dated Daily Note. |
| Daily Notes | append_daily_note_markdown |
Add Markdown to today's or a dated Daily Note. |
| Media uploads | upload_files |
Stream one or more local files into Capacities media objects with readback verification. |
| Media uploads | manage_upload_job |
Inspect, wait for, or cancel an in-process background upload job. |
upload_files accepts absolute local file paths only; relative paths, directories, and empty files are rejected. Background upload jobs live only in the current MCP process and retain terminal status for a limited time.
Supported environment variables and configuration options
| Setting | Required | Default | Description |
|---|---|---|---|
CAPACITIES_API_TOKEN |
Yes, unless every call supplies apiToken |
— | One API key or a comma/semicolon-separated key pool. The keys must belong to the same space and use the same permissions. |
CAPACITIES_MCP_READBACK |
No | true |
Set to false to disable ordinary mutation readback. Rollback verification remains enabled. Accepted values: true, false. |
Per-call configuration is exposed through the optional apiToken argument on
all tools. It overrides CAPACITIES_API_TOKEN for that call and accepts the
same key-pool syntax.
⚙️ API key pool and rate limits
CAPACITIES_API_TOKEN and the per-call apiToken option accept multiple API
keys separated by , or ;.
Keys are selected with round-robin scheduling independently for each endpoint,
skipping keys currently marked as rate-limited.
All keys in one pool must belong to the same space and have the same permissions. At least 3 keys are recommended; use separate MCP server instances for multiple spaces.
On 429, the pool immediately fails over to another key that is not rate
limited. The error is returned only when all keys are rate limited. No waiting,
exponential backoff, or cross-pool retry is performed.
🚅 Concurrency and safe writes
Reads for the same object can run concurrently. A mutation obtains an exclusive object lock, so it waits for active reads and other writes; reads arriving during a write wait as well. Operations for different objects proceed concurrently. This protects the full read → mutate → readback lifecycle against last-write-wins races. The current Capacities SDK does not expose request cancellation/timeout configuration, so this server does not use an unsafe client-side timeout race for mutations.
🛠️ Development
bun run lint
bun run typecheck
bun test
bun run verify:tools
bun run build
The final local release gate is:
npm run release:check
npm publish --dry-run --access public
npm publish --dry-run validates the package without publishing it. Actual
publishing additionally requires an authenticated npm account.
Live API tests are available through bun run test:live when a dedicated test token is configured.
📚 Documentation
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.