OnlyMCP

Delete one agent. It moves to the trash and can be restored in the dashboard until the trash empties (after 3, 7 or 30 days).

Create an agent or replace an existing one. description and body are required; model and tools are optional and keep their previous value when left out. Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

Call a tool of a connected server by its name "__", only for a server connected after your client last listed its tools, whose tools it does not list yet; a tool of a server your client has listed is refused here. Same as calling the tool directly: the same rights, the same count against your plan. Answers what the tool answers.

Connect a remote MCP server. Only http and sse servers - a stdio server runs as a child process on a machine and cannot be started on somebody else's behalf. The server is probed straight away and the result reported. Passing a secret here puts it into this conversation; leave it out and set it in the dashboard if that matters. Docs: https://only-mcp.com/docs/mcp-servers#how-a-server-signs-in

Disconnect a remote MCP server and forget its stored credential. Its tools disappear from this gateway.

List what one connected server offers - tools, resources and prompts - and whether each one is switched on for this gateway.

Show what one tool of a connected server takes: its description and its input schema, by its name "__". Only for a server connected after your client last listed its tools, whose tools it does not list yet.

Switch one tool of a connected server on or off. A tool that is off costs no room in any client's tool list.

Change a connected MCP server. Only the fields you pass are touched. Changing url, transport, authType or the secret triggers a fresh probe. Docs: https://only-mcp.com/docs/mcp-servers#how-a-server-signs-in

Load the full definition of one agent by name.

Load the full body of one skill by name.

Read one companion file of a skill. Binary files come back base64 encoded; the encoding is on the first line. Docs: https://only-mcp.com/docs/import-and-sync#keep-clients-in-step

List every agent name available to this account, private and team.

List the MCP servers connected through this OnlyMCP account, switched-off ones included, with the tool-name prefix each one uses.

List the paths of every memory file in this account, private and team, sorted by path.

List the companion files of one skill, with their size in bytes.

List every skill name available to this account, private and team.

List the teams this account is in, one per line: the team id, its name and your role in it.

Append text to one memory file, after a newline when the file does not end with one. A missing file is created. Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

Delete one memory file. It moves to the trash and can be restored in the dashboard until the trash empties (after 3, 7 or 30 days). Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

The graph around one memory file, up to three hops. Depth 1: the files, skills, agents and servers it links to or is linked from, its folder, type, project and tags, and links that land on nothing (→ links to (missing) name). From depth 2 on only links out of a file reached by a link are followed: a backlink, folder, type, project or tag is listed at depth 1 and never walked through. Pick the depth yourself; 1 is the default and usually enough, 2-3 follow a chain of links further, the files most of the previous depth links to first. Grouped by depth; each line names the edge (→ links to, ← linked from, → has type ...) and, from depth 2, the node it was reached through. types, folder and project narrow what is listed, not the walk.

The [[links]] in one memory file with the kind each lands on (memory, skill, agent, mcp), the ones that point at nothing yet, and what links to the file. Docs: https://only-mcp.com/docs/what-is-onlymcp#links-between-memory-skills-agents-and-servers

Move or rename one memory file; one file per call, a folder is not moved as a whole. Meant for reorganising without breaking links: by default every [[link]] that would no longer find the file is rewritten in the files that hold one, whatever its form (a name, a path, an alias, a heading), and the file keeps its history, which a new copy would not. A file you may only read is left alone. Answers "Moved to " with how many files had links rewritten and how many were left alone. Set updateLinks to false only when old links should stay as written.

Read one memory file by path, whole or a range of its lines: from is the first line (counted from 1) and lines how many; the answer then names the total number of lines of the file. Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

Full-text search across memory, skills and agents. A plain word matches as a prefix (kal finds kaleidoscope), a file whose path or name contains the query is a hit and comes first, and a name with a small typo is found where the database allows it. Quotes and -word switch to exact search. Between equally good matches the more recently changed one ranks first, and a file marked superseded_by in its frontmatter comes after the current ones, with → superseded by . type, project and tag narrow the search to memory files whose frontmatter says so (metadata.type, metadata.project or the sol-{project}-/proj-{project} file name, metadata.tags). Returns the best matches, each with the day it last changed and a short snippet; the snippet only points at a hit, so read the hit itself before answering from it. expand=links adds, under each of the first five, the files, skills and agents it links to and that link to it (← marks those).

Show how this account's memory is organised, as an indented tree: every folder, empty ones included, with the names of the files in each, and the team an item belongs to. Meant for seeing the layout before choosing where a new file goes; it returns names only, no file contents, each file with its frontmatter type in brackets ([solution]). scope narrows it to personal files or to one team, folder to one folder and what is under it.

Create or replace one memory file with the full text given. Meant for a new file or a complete rewrite, not for adding a few lines to the end of a file. Replacing overwrites the whole text; recent versions stay restorable from the file's history in the dashboard. Folders in the path need not exist. A new file under a team's folder is created in that team. With ifMatch the write is refused when the file changed since it was read. When a note replaces an older one, add superseded_by: to the old file's frontmatter: search then lists it after the current one. Answers "Saved ". Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

Delete one skill with every file that belongs to it. It moves to the trash and can be restored in the dashboard until the trash empties (after 3, 7 or 30 days).

Delete one companion file of a skill; the skill and its other files stay. Meant for removing a script or reference the skill no longer uses. The file is removed at once, without a trash, and the skill's history records the deletion. Answers "Deleted /", or says that no such file was there. Docs: https://only-mcp.com/docs/import-and-sync#keep-clients-in-step

Create or replace one companion file of a skill: a script, template or reference that sits next to its SKILL.md. Meant for files beside the skill body; the SKILL.md itself is written with the skill. Text goes in as utf8, a binary file as base64. Replacing overwrites the whole file. A file may hold 2 MB, a skill 300 files and 24 MB in all. Answers "Saved / ( bytes)". Docs: https://only-mcp.com/docs/import-and-sync#keep-clients-in-step

Create a skill or replace the body of an existing one. The body is the whole SKILL.md, frontmatter included. Docs: https://only-mcp.com/docs/what-is-onlymcp#what-sits-behind-the-one-address

Take an item out of a team and back under its owner. Name one of your own items, or pass the id of any item of the team if you may delete that kind of content there. The item keeps its owner, and the return is never refused on a quota.

Move one of your own memory files, skills or agents into a team, so its members can use it. You need the right to create that kind of content in the team, the team must not hold an item of that name already, and a frozen team takes nothing new.