Agent & MCP Tools

Scholardo runs a local MCP server for every open project. Any MCP-compatible agent that connects can search your library, read annotations, manage todos, and fetch papers — without leaving the terminal.

Supported agents

Agent	Notes
Claude Code	Recommended. First-class support.
Codex CLI	Fully supported.
pi	Fully supported, native MCP bridge.
DeepSeek / Kimi / GLM, etc.	Via the Claude Code carrier.
Any MCP-compatible agent	Configure manually.

See Which AI should I choose? for the tradeoffs.

Connecting

Agents launched from Scholardo's right-hand pane connect automatically to the current project's MCP server — no configuration needed. This is one of the main reasons to run your agent inside Scholardo rather than in a separate terminal.

Each project gets its own MCP server and socket, managed by the app. Switch projects and the agent sees the new project's contents.

Verify the connection:

scholardo-mcp list

This lists every tool available in the current project. To inspect one tool's arguments:

scholardo-mcp describe search

Tool naming

Agents see tools under the MCP namespace prefix: mcp__scholardo__search. The names below are without the prefix — the same names scholardo-mcp describe takes.

Available tools

Search & retrieval

Tool	Purpose
search	Library-wide search (lexical / dense / hybrid), see Semantic Search
get_item	Fetch one item by ID
batch_get_items	Fetch several at once
list_sources	List the project's sources
list_refs	List reference entries
list_notes	List notes

Annotations (read-only)

Tool	Purpose
list_annotations	List all annotations on a document
get_annotation	Fetch one by ID
search_annotations	Search by text or tag

Annotations are read-only to the agent

There is no write tool, by design. The agent can read all your markup to summarize and compare, but cannot create or modify it — marking up stays yours. See Annotations.

Notes & memory

Tool	Purpose
create_note / update_note	Create / update notes
save_memory / read_memory / update_memory / list_memory	Read and write the agent's long-term memory entries

Todos

Tool	Purpose
list_todos	List todos
create_todo	Create one
toggle_todo	Toggle done state
set_reminder	Set a reminder

Paper discovery & retrieval

Tool	Purpose
search_papers	Federated paper-metadata search
search_openalex	Search OpenAlex
lookup_unpaywall	Look up open-access full text
download_pdf	Download full text into Docs
pairing_audit	Report gaps between Docs and Refs
queue_add	Add a paper to the reading queue

Feeds

add_feed · remove_feed · refresh_feed · set_feed_filter

See Feeds & Digests.

Active state ("what is the user looking at")

Tool	Purpose
active_context	Current project / file / tab
preview_text	Text in the active preview
web_page_context	The in-app browser's current page (requires explicit sharing)
active_role	Current agent role
context_status	Context status

Other

Tool	Purpose
get_tags / set_tags	Read / write a file's Finder tags
get_writing_context / set_writing_context	Read / write the writing brief
get_rules	Read the active agent rules
get_settings	Read relevant settings
scholardo_list_agent_assets	List available skills / subagents / workflows
scholardo_slide_check	Check a slide deck for content overflow
ping	Connectivity check

Why two long names

Most tools use short names — MCP already namespaces calls by server, so a scholardo_ prefix is redundant. scholardo_list_agent_assets and scholardo_slide_check have not been folded into that rename yet.

Live resources

Beyond tools, Scholardo exposes MCP resources. Clients that speak the resource protocol (such as Claude Code) can attach them to context without spending a tool-call slot:

Resource	Contents
scholardo://active/context	Active project and file
scholardo://active/preview-text	Text in the active preview
scholardo://active/web-page-context	The in-app browser's current page

So when you say "summarize these and find related work," the agent knows what "these" refers to.

Calling from the command line

scholardo-mcp doubles as a CLI, letting tools without native MCP support reach Scholardo indirectly:

scholardo-mcp call search '{"query": "solid electrolyte interphase impedance"}'

Other subcommands: list · describe <tool> · instructions

Too many tools?

Scholardo adapts how tools are exposed to each agent's capabilities — models with tight context windows get tool definitions loaded on demand, roomier ones get them up front. You do not need to configure this.

See also

• Semantic Search — the retrieval behind search
• Annotations — why annotations are read-only