Rithmo
Server Details
Rithmo provides resolved business context for AI agents, helping prevent them from acting on stale, conflicting, or superseded information. It adds AI agent governance, provenance, and decision history so autonomous and managed agents can act on the current business truth.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 23 tools
Most tools target clearly distinct governance steps: start/verify actions, manage approvals, query decisions, and report outcomes. The main overlap is among agent-configuration tools (declare_my_setup, update_agent, set_agent_governance, register_my_agent), though their descriptions distinguish purpose, instructions, governance, and creation.
The set is overwhelmingly consistent snake_case verb_noun (check_premises, request_approval, report_action_outcome, list_my_agents). Minor deviations like whoami and current_source_pointer are readable but do not follow the verb_noun pattern.
At 23 tools this is somewhat heavy, but the server is a multi-stage governance and approval layer with setup, run lifecycle, pre-action checks, approvals, decision polling, and reporting. Each tool appears to cover a distinct protocol step rather than duplicating surface area.
The surface covers the full governance lifecycle: setup reporting, run start/end, pre-action check and verification, hold resolution, approvals, decision lookup and change polling, and agent management. Minor gaps exist, such as no explicit agent retirement/deletion operation, but core workflows are complete.
Available Tools
23 toolsanswer_setupReport your person's setup answersARead-onlyInspect
Call after asking your person the setup_chat questions start_agent_run gave you, in a run they started. run: from start_agent_run. report: setup_chat.report, exactly (null when it was null). confirmed: true only when they said yes to your one-sentence account of what you do. said: their answer to that question, exactly. name: what they want you called. purpose: the sentence you showed them (or their correction). systems: any systems they added. documents: links they gave you. Rithmo records it as your report.
Inputs: run (required), said (required), confirmed (required), name, +6 more.
approve_from_chatRecord an approval given in the chatARead-onlyInspect
Call only when the person you act for wrote the approval word to you in this chat about a text you sent for approval (request_approval). run: from start_agent_run, started_by person. content_hash: "sha256:" + hex SHA-256 of the exact text they approved. action_fields: the SAME action_fields you sent with request_approval (none if you sent none) -- show them to the person in the chat with the text, since the approval stands only for them. said: their message, exactly. Rithmo records it as your report; if it refuses, wait for their reply where Rithmo asked. Then act on it as on any approval.
Inputs: run (required), said (required), approval_id (required), content_hash (required), +2 more.
check_premisesCheck the premises an action relies onARead-onlyInspect
Call it before EVERY action that changes or sends anything in any system, listed in get_my_setup or not (D-053). Stage 1 of the pre-action guard. Given the action you are about to take and the PINNED subject ids it relies on, returns allow/hold/deny per premise plus a short-lived lease for each cleared one. Takes subject ids, never a natural-language question — retrieval wording cannot change the verdict. To check the DOCUMENT you work from, omit subject_id and pass the document from get_my_setup as relied_on with content_hash ("sha256:" + hex SHA-256 of the exact text you read): allow only when that is the version its owner confirmed. Do the work, then call verify_premises with these leases IMMEDIATELY before acting. Fails closed: an unknown pin, an unreadable record or a premise that cannot be leased all return hold, never allow. Give Rithmo time to answer: wait at least 10 seconds for this call. No answer in that time is not a yes: do not act, as for a hold.
Inputs: premises (required), intended_action (required), run, surface, +2 more.
current_source_pointerGet the source backing the current decisionARead-onlyIdempotentInspect
Given a PINNED subject id, returns the provider-native identity of the external resource that currently backs that subject's answer — ready to pass straight back as relied_on on check_premises. Use it after check_premises held you for a superseded source: retrieve the current answer, call this for the current source, then re-check and verify. Returns identity ONLY (provider + resource id, e.g. a Google Drive file id or ":"; supported: slack, google_drive, github) — never document text, message content or summaries. Fails closed: when no declarable source exists it returns result="no_source_pointer" with a typed code and safe_to_declare=false. Never declare a source this tool did not return.
Inputs: subject_id (required).
declare_my_setupReport your own instructions and toolsARead-onlyInspect
Tell Rithmo what you are actually instructed to do and which tools and knowledge you have in your own platform, so your page in Rithmo shows you as you are and flags when that changes. Send your instructions exactly as you were given them (a Claude Code agent: your agent file's body, word for word): the complete text, every line, never summarized, shortened or reworded, however long. The answer gives the characters Rithmo recorded and how many secrets it removed (redactions); if it recorded fewer characters and removed none, send them again in full. Never send credentials, tokens or secrets. Call it at the start of every run, with agent_key; unchanged reports are cheap. If Rithmo has an instructions file for you, send only your tools (no instructions), every run. If Rithmo has no instructions file for you, this report is required: check_premises holds you when you have not reported in the last 15 minutes, or when your instructions differ from the ones your owner confirmed (until they accept the change). Otherwise a person must first allow your agent to report. A host adapter sends protocol "rithmo.host_baseline.v1" with host_agent, runtime, tools_mode, permissions, mcp_connections, resources and hooks; anything it cannot read is sent as "unknown" (or left out), never as empty. It grants nothing.
Inputs: source_format (required), hooks, scope, title, +12 more.
get_approvalRead an approvalBRead-onlyIdempotentInspect
The status of an approval you asked for (pending, approved, superseded, used) and the approver's replies to it, read from where Rithmo asked them, oldest first. A reply other than the approval word is theirs to you: what to change, or a question.
Inputs: approval_id (required), agent_key.
get_hold_statusGet the status of your held callARead-onlyIdempotentInspect
Given a hold_id that check_premises or verify_premises returned to you, says whether a person has dealt with it yet (open, resolved or withdrawn). It never authorizes anything: while open, do not act; once closed, call check_premises again with resumes_hold set to this hold_id and act only on an allow you confirm with verify_premises. Only your own holds are visible. Read-only.
Inputs: hold_id (required), run.
get_my_setupGet my setupARead-onlyIdempotentInspect
Returns the setup your organization's admin gave this agent: its instructions, the documents it may rely on (each with the relied_on identity to declare to check_premises; an action must rely on the resources it works from, and on every one marked required for its kind of work, or it holds), the tools it may use, and actions: approval rules -- actions that also need a person's approval, with the exact intended_action to pass. before_every_action: check EVERY action with Rithmo before taking it, listed or not; actions never narrows that. Call it at the start of each run and work only from what it returns. Each agent also carries its config_revision — pass it to update_agent as expected_revision to edit its purpose WITHOUT starting a run — and can_update (whether this connection is permitted to edit at all). If your connection serves several agents, pass the agent_key from list_my_agents to select one; omit it to get them all. report_instructions.required: Rithmo knows your instructions only from what you report, so call declare_my_setup with them, complete, now, before your other work; Rithmo holds the actions it checks until it has them. live: false: a person has not put the agent live in Rithmo yet; do what until_live says and nothing else. blocked: true: a person must review your instructions first; do what until_unblocked says and nothing else. Read-only — this call never starts a run.
Inputs: agent_key.
get_rithmo_protocolGet the Rithmo protocolARead-onlyIdempotentInspect
Returns the Rithmo protocol: the steps you follow with Rithmo every run (start, a check before every action, approvals, end of run), its version and its content_hash. start_agent_run returns the same protocol in its answer. Takes no arguments.
Inputs: no named parameters.
list_decision_changesList decision changesARead-onlyIdempotentInspect
Drain decisions whose state changed, since a cursor. Call with no since to start from now (returns an empty set + the current cursor); then call repeatedly with the returned cursor to get changes in order. Includes reversals (dropped/superseded). Read-only; poll this on a schedule.
Inputs: limit, since.
list_my_agentsList my agentsARead-onlyIdempotentInspect
Call get_rithmo_protocol first, every run. Lists the agent profiles you own on this Rithmo connection — their agent_key, purpose and owner. One connection can serve several agents; call this to discover which to act as, then pass the chosen agent_key to get_my_setup (and to check_premises/verify_premises). disconnected_agents lists any a person disconnected from Rithmo: Rithmo refuses their calls until they are reconnected in Rithmo. Read-only; takes no arguments.
Inputs: no named parameters.
note_to_approverTell the approverBRead-onlyInspect
Post a short note to the approver in an approval request's thread, from Rithmo's app: what happened to the action, or an answer to their question. It never counts as an approval.
Inputs: text (required), approval_id (required), agent_key.
query_decisionQuery a decisionARead-onlyIdempotentInspect
Look up the current authoritative decision for a natural-language question, before an execution step. Returns the answer, its owner, its source, and status; result="found_record" when the subject record holds the answer; or result="no_record" with a typed refusal (record: closed with its disposition, no_record with a reason, or not_observed when the surface is not connected) carrying premise_check.safe_to_proceed=false — do NOT act on the premise when the record refuses. Read-only; the record is Rithmo's, the action is yours.
Inputs: question (required), area, type, owner, +1 more.
register_my_agentRegister another of your agentsAInspect
Create ANOTHER agent profile on THIS connection — no new sign-in — so you can run several agents (each with its own instructions) through one authorization. Requires the mcp:agents:write permission. Pass a name and a unique request_key (retrying the same key is idempotent: it returns the same profile, never a duplicate). Returns the new agent_key; then call get_my_setup + update_agent to set its purpose (its owner sets its instructions). The profile is yours, on this installation only — it grants no new tools or sources and never touches another member's agent. Refused (not_authorized) if you lack the permission or this connection is not a hosted installation you own.
Inputs: name (required), request_key (required).
report_action_outcomeReport what you did after an allowed actionARead-onlyInspect
Call AFTER acting on an allow from verify_premises, once per lease: report whether you executed the action, abandoned it, or tried and failed (optionally with the external system's identifier for it, never content). Requires the mcp:report scope. Accepted only for a lease this same connection used in verify_premises, with the same intended_action. Recorded as evidence of what you report; it is NOT proof of enforcement and authorizes nothing.
Inputs: lease (required), outcome (required), intended_action (required), run, +1 more.
report_agent_runReport a run you finishedARead-onlyInspect
Call once when a run ends: its start and end time (ISO 8601), whether it completed or failed, how it ended (outcome: result -- it ended with a usable result; no_result -- it ended without one, or inconclusive; failed), the model, and the tokens it used (input, output, cache writes, cache reads; fast-mode calls separately in fast_usage; if you cannot see your token counts, estimate them -- the characters of everything you read and wrote, divided by 4 -- and set usage_estimated: true beside usage (not inside it); never send zeros for them; always say the model you ran on, exactly as your host names it: it prices the tokens, and Rithmo checks it against the models your owner set and alerts them to a different one), and optionally the same tokens by model (usage_by_model: up to 20 { model, speed: standard | fast, long_context, usage }) so each model is priced at its own rate. run_ref is your own id for the run; reporting the same run again records nothing new. Requires the mcp:report scope. Rithmo prices the run itself; the run shows on the agent's page as reported by the agent. Optionally say what the run was asked (task, a few words), how it ended (result, a few words) and what it read (read: up to 50 { kind: file | url | tool, ref }) and the work behind it (work: up to 10 { system, kind: repo | branch | commit | pull_request | ticket, label, url (https) }) and where its tokens went (breakdown: { kind: instructions | research | editing | testing | sharing | tools | delegating | thinking | other, usage }, the same tokens split by kind of work, adding up to usage plus fast_usage for every token type; one that does not is left out) and where you left your result for a person to see (delivered_to: { where, in a few words, e.g. "ChatGPT chat" or "Slack #marketing"; url (https) of it when you have one }), which the agent's page shows so its owner knows where to look. These are shown as your own account of the run; they grant nothing. Always say how many actions the run took that changed or sent anything in any system (actions_taken; 0 when none): Rithmo compares it with the checks you made, and a run that acted without checking is blocked until a person reviews it.
Inputs: status (required), run_ref (required), ended_at (required), read, +14 more.
request_approvalAsk for approvalARead-onlyInspect
Ask the person you act for, from the authority map, to approve one action. Rithmo sends them text exactly, through its own app in a connected system, and reads their reply itself: they approve by replying approval_word. To act on it, rely on it in check_premises: relied_on [{ provider: "approval", resource_id: approval_id, content_hash: "sha256:" + hex SHA-256 of the same text }]. It allows only after they approved that exact text, and only once. replaces: the approval_id of an earlier version this text replaces (sent in the same thread; the earlier one can no longer be used). where: the channel the team uses for this work (e.g. "#marketing"); a direct message to the approver when absent. Only the approver's own reply counts, wherever it is asked. action_fields: the action's details (recipient, amount, target...), shown to the approver under the text; the approval then allows only a check_premises with the same action_fields. in_chat_run: the run (from start_agent_run, started_by person) when the approver is in this chat with you and sees the text here: Rithmo then sends nothing anywhere, and only their approval word to you here, recorded with approve_from_chat, approves it. in_chat: true when your own tool, with no run, shows the text in your chat: the same, nothing is sent.
Inputs: text (required), approval_word (required), intended_action (required), title, +6 more.
rithmo_actionCall a Rithmo stepARead-onlyInspect
Calls the Rithmo tool named by action with args, exactly as calling that tool by name would. Use it for any Rithmo step your Rithmo protocol names that your host does not list as a tool. The answer is that tool's answer.
Inputs: action (required), args.
set_agent_governanceSwitch Rithmo on or off for this agentBRead-onlyInspect
Only when the person who owns this agent asks, in their own words, to turn Rithmo on or off for it: on true or false, said their words, run the run they started. Refused off when their company requires Rithmo for this agent.
Inputs: on (required), agent_key (required), run, said.
start_agent_runStart a run as one of your agentsARead-onlyInspect
Call it first, every run, before anything else. When the answer has setup_chat, your person has not set you up with Rithmo yet: show them setup_chat.message as it is (fill in its [brackets]) before their task, wait for their answers, report them with answer_setup, then do their task. Begins a run as the agent you name (its agent_key from list_my_agents; or, when you do not know your key, your own name as agent_name: Rithmo finds you, or signs you up on this connection the first time, and answers with your agent_key; with neither, the only agent on this connection) and returns a run handle plus the config_revision it was issued against. Pass run to check_premises and verify_premises so Rithmo knows which of your agents is acting. before_every_action: the rule for the whole run -- check every action with Rithmo before taking it, whatever any list says. Refused if you do not own that agent (or it was retired). live: false: do what until_live says and nothing else; blocked: true: do what until_unblocked says and nothing else. started_by: person when someone started this run by writing to you, schedule when it began from your schedule, trigger when Rithmo started it because an approval was answered. protocol_version: the Rithmo protocol version you follow (the one your last protocol answer named). The answer carries protocol, the Rithmo protocol as it stands now, and instructions, your own instructions as their file stands now: follow both for the whole run, in place of any older copy.
Inputs: agent_key, agent_name, started_by, protocol_version.
update_agentUpdate one of your agentsAInspect
Set the purpose of an agent you own (its agent_key from list_my_agents). Requires the mcp:agents:write permission AND that you own that agent on this connection — another agent's setup is never writable. Pass purpose ("" clears it). You MUST pass expected_revision — the config_revision you read from get_my_setup — so a concurrent edit is detected: if the agent's configuration changed since you read it, the write is refused (conflict) and nothing is written; re-read get_my_setup and try again. Returns the new setup and its config_revision; a run you already started keeps its own pinned revision until you start a new one. Refused (not_authorized) if you lack the permission or do not own the agent. An agent never changes its own instructions: a write that includes instructions is refused (not_agent_writable) and nothing is written. Your owner sets them on your page in Rithmo, or an org admin on the Control Board; report what you run with through declare_my_setup.
Inputs: agent_key (required), expected_revision (required), purpose, instructions.
verify_premisesVerify leased premises before actingARead-onlyInspect
Stage 2 of the pre-action guard. Re-checks the leases from check_premises against the live record and returns allow only when every one still holds, deny when the record now says the action must not proceed (a decision was closed), and hold otherwise. Catches a record that MOVED while you were preparing — compared by version, so an A to B back to A revision is still detected. Call this immediately before the external action, not earlier. Fails closed on expiry, version movement, a changed intended action, a bad signature or an unreadable record.
Inputs: leases (required), intended_action (required), run, action_fields.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceRemote MCP server that lets AI assistants read tracked competitive intelligence — competitor news, reviews, website changes, social posts and ads — and act on it by managing projects, competitors, battle cards, reports and canvases. Every answer carries a source and a date, and write actions only publish when explicitly shared.MIT
- AlicenseCqualityCmaintenanceCompetitor Monitor AI - MCP server providing AI-powered tools and automation by MEOK AI Labs112 npm37 PyPIMIT
- AlicenseAqualityDmaintenanceEnables brand visibility monitoring across major AI platforms like ChatGPT, Claude, Gemini, and Perplexity. It allows users to track visibility scores, analyze competitor data, and receive actionable insights to improve AI-generated brand recommendations.1611 npm1MIT

industrylens-mcpofficial
AlicenseNot gradedqualityCmaintenanceBrowse IndustryLens's published competitive-intelligence reports and head-to-head competitor comparisons from any AI agent — real, source-backed data.MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.