MCP Tools

dtoolkit exposes its functionality through the Model Context Protocol (MCP), allowing AI coding agents to interact with memory, project management, and observability services using structured tool calls. Each service runs an MCP server alongside its REST API on the same port.

Tip

All MCP servers use Bearer token authentication. Generate API keys with the keys CLI subcommand for each service (e.g., dbrain keys create).

dbrain MCP Tools

dbrain is the persistent memory server. Its MCP tools let agents store, search, and manage memories organized as entities and facts with automatic tiering (hot/warm/cold).

Port: 7878 (shared with REST API)

Tool	Description	Parameters
recall	Search memories across all entities. When federation is enabled, results include both local and connected shared brain memories.	query (string, required), limit (number), federated (boolean)
remember	Save a single atomic fact about an entity. Creates the entity if it does not exist.	entity (string, required), fact (string, required), source (string)
get_entity	Retrieve full context for a specific entity, including all associated facts and metadata.	name (string, required)
list_entities	List entities filtered by category or type. Returns name, category, type, and summary for each.	category (string), type (string)
create_entity	Create a new entity with metadata. Use remember afterwards to attach facts.	name (string, required), category (string, required), type (string, required), summary (string)
bump	Touch a fact to reset its decay timer and keep it in the hot tier.	factId (string, required)
log	Store conversation messages at natural breakpoints for later review.	messages (array, required)
overview	Return brain statistics: entity count, fact count, tier distribution, connection status.	—
share	Push a fact to a connected shared brain. Only available when the brain has active connections.	factId (string, required)
compact	Deduplicate facts and recalculate memory tiers. Admin-only operation.	—

Caution

The compact tool is restricted to admin API keys. Running it on a large brain may take several seconds. Avoid calling it during high-throughput sessions.

Memory tiers

Facts in dbrain are automatically organized into three tiers based on recency and access frequency:

Tier	Behavior
Hot	Recently accessed or bumped. Prioritized in recall results.
Warm	Accessed within a moderate window. Included in recall but ranked lower.
Cold	Stale facts. Still searchable but deprioritized. Candidates for compaction.

Use bump to promote important facts back to hot tier. The compact tool deduplicates redundant facts and recalculates tiers across the entire brain.

Federation

When a personal brain is connected to a shared brain (via dbrain connect), the recall tool automatically federates queries across both. Use share to explicitly push a fact from your personal brain to the shared brain.

dwork MCP Tools

dwork is the AI-native project manager. Its 21 MCP tools plus 5 graph tools cover project CRUD, task management, documentation, search, and code intelligence.

Port: 7881 (shared with REST API)

Project management

Tool	Description	Parameters
create_project	Scaffold a new project with directory structure and initial BACKLOG.md.	name (string, required), description (string), template (string)
get_project	Retrieve project metadata, task summary, and doc count.	projectId (string, required)
list_projects	List all registered projects with status summaries.	—
update_project	Update project metadata (name, description, status).	projectId (string, required), name (string), description (string), status (string)

Task management

Tasks are stored in each project’s BACKLOG.md file. The SQLite FTS5 index is derived from these Markdown files.

Tool	Description	Parameters
add_task	Add a new task to a project’s BACKLOG.md.	projectId (string, required), title (string, required), description (string), priority (string), status (string)
get_tasks	List tasks for a project, optionally filtered by status or priority.	projectId (string, required), status (string), priority (string)
update_task	Update task fields (title, description, status, priority).	projectId (string, required), taskId (string, required), title (string), description (string), status (string), priority (string)

Documentation

Tool	Description	Parameters
add_doc	Add a numbered Markdown document to a project.	projectId (string, required), title (string, required), content (string, required)
get_doc	Retrieve a specific document by ID.	projectId (string, required), docId (string, required)
get_docs	List all documents for a project.	projectId (string, required)
update_doc	Update document title or content.	projectId (string, required), docId (string, required), title (string), content (string)

Search and overview

Tool	Description	Parameters
search	Full-text search across all projects, tasks, and docs using FTS5 OR matching.	query (string, required), projectId (string)
overview	Global statistics: project count, task breakdown by status, recent activity.	—
sync	Re-index Markdown files into SQLite. Run after external edits to BACKLOG.md.	projectId (string)

Planning

Tool	Description	Parameters
get_roadmap	Retrieve the project roadmap derived from task priorities and statuses.	projectId (string, required)
what_to_do_next	AI-prioritized suggestion of the next task to work on, based on status, priority, and dependencies.	projectId (string)

Code graph tools

These tools are powered by @dtoolkit/codegraph-sdk and provide semantic code intelligence over the project’s codebase.

Tool	Description	Parameters
graph_search	Semantic search over code symbols (functions, classes, types) in the project’s knowledge graph.	query (string, required), projectId (string), kind (string)
graph_stats	Statistics about the code graph: symbol counts by kind, file counts, edge counts.	projectId (string, required)
graph_trace	Trace the dependency chain of a symbol — what it depends on and what depends on it.	projectId (string, required), symbol (string, required), direction (string)
graph_impact	Analyze the impact radius of changing a symbol — all transitive dependents.	projectId (string, required), symbol (string, required)
graph_context	Retrieve full context for a symbol: definition, references, documentation, and surrounding code.	projectId (string, required), symbol (string, required)

Tip

The graph tools require an indexed codebase. Run dwork sync after cloning a repository or making significant structural changes to rebuild the code graph.

dops MCP Tools

dops provides agent observability. Its MCP tools allow agents and dashboards to log sessions and query usage metrics.

Port: 7883 (shared with REST API)

Tool	Description	Parameters
log_session	Ingest a completed session with token counts, cost, tool usage, and outcome.	sessionId (string, required), provider (string, required), tokens (object), cost (number), tools (array), success (boolean), error (string)
get_stats	Retrieve aggregated statistics: total sessions, tokens, cost, success rate, by provider and time range.	provider (string), from (string), to (string)
get_costs	Cost breakdown by provider, model, and time period.	provider (string), from (string), to (string), groupBy (string)
get_sessions	List recorded sessions with filtering and pagination.	provider (string), from (string), to (string), limit (number), offset (number)
health	Service health check. Returns status and uptime.	—

Tool discovery

MCP clients discover available tools automatically when connecting to a dtoolkit service. Each service exposes its tool list via the standard MCP tools/list method.

To connect an AI coding agent to dtoolkit services, use dcontext:

terminal

dcontext init
dcontext install

This configures the agent’s native hook system (e.g., Claude Code’s settings.json, Gemini CLI’s settings.json) to connect to running dtoolkit MCP servers.

Manual MCP connection

If you prefer manual configuration, point your MCP client at the service URL:

terminal

# dbrain MCP endpoint
http://localhost:7878/mcp

# dwork MCP endpoint
http://localhost:7881/mcp

# dops MCP endpoint
http://localhost:7883/mcp

All endpoints accept standard MCP JSON-RPC messages over HTTP with SSE streaming for server-initiated messages.

Cross-service patterns

All dtoolkit MCP services share these conventions:

Pattern	Detail
Authentication	Bearer token via Authorization header. Generate keys with <service> keys create.
Transport	MCP over HTTP on the same port as the REST API.
Error format	Standard MCP error responses with code and message fields.
Idempotency	Read operations are safe to retry. Write operations use unique IDs to prevent duplicates.
Permissions	Some tools (e.g., compact) require admin-level API keys.