Build and manage AI agents on Keiki through MCP. Create agents, update their models and system prompts, inspect conversation histories, monitor engagement analytics and quality reviews, and evaluate agent performance using test cases and evaluation runs.
Hermes is Nous Research's command-line agent. MCP servers are declared in its YAML config and reloaded without a restart.
This guide connects Keiki MCP to Hermes through MewCP, so Hermes can use Keiki MCP's 2 tools, like search and execute, without your Keiki MCP credentials ever reaching the app.
Keiki MCP signs in through managed OAuth. MewCP stores the token encrypted, refreshes it before it expires and injects it only at call time. Your agent never holds it, so it can never leak it.
More apps
TinyFish MCP for running AI-driven browser automation tasks, deep web research, persistent browser login profiles, remote CDP browser sessions, and agentic wallet payments.
TypeSafe's Jev model for typed decisions — ask yes/no, multiple-choice or rubric-score questions about a block of state and get calibrated probabilities back instead of prose. Built for classification, routing and scoring steps inside a workflow
Connect to external APIs, webhooks and internet services through standard HTTP requests. Build integrations with virtually any platform that exposes a web-based interface.
Turn your AI agent into a creative studio The world's best image, video, audio and 3D models, inside the AI assistant you already use.
Other clients
Keiki MCP works the same way in every client below. Pick yours for its own guide.
Free to start.
Scope access to the exact Keiki MCP tools and accounts you allow, work and personal side by side. Anything outside that scope is refused at call time, even when the model asks for it.
Each call is logged: what ran, on which account, and whether it worked. Failures come back structured, so the agent knows whether to retry instead of guessing.
Every app sits behind one secure MCP URL, and there is no server for you to deploy or keep running. Add an integration in the dashboard and it is live on the next call, with no config to change and no new credentials to wire up. Tools load only when a task needs them, so your agent stays fast and focused however many apps you connect.
Edit config.yaml in your Hermes config directory (~/.hermes).
Put this under the top-level "mcp_servers:" key.
Run /reload-mcp in Hermes, or start a fresh session. Sign in when the browser window opens.
Start a new chat and ask: "List the tools you can use from MewCP." You should see your connected apps.
Find Keiki API calls by words (every word must match), by exact path (`agents.get_by_id`), or by group. Returns each call's TypeScript signature as used inside execute. An empty query lists the group. Groups: keiki.agent_goals (1), keiki.agent_groups (12), keiki.agent_templates (1), keiki.agentmail (2), keiki.agents (86), keiki.approvals (3), keiki.audit (2), keiki.auto (1), keiki.axiom (1), keiki.billing (18), keiki.blog (2), keiki.capability_tools (1), keiki.channel_access_requests (2), keiki.conversations (14), keiki.dashboard (1), keiki.evals (16), keiki.identities (53), keiki.insights (1), keiki.invitations (4), keiki.keys (5), keiki.lines (10), keiki.live (1), keiki.loops (1), keiki.marketplace (5), keiki.mcp (3), keiki.mcp_clients (2), keiki.members (4), keiki.models (1), keiki.notifications (2), keiki.org_mcp_integrations (5), keiki.orgs (6), keiki.outcomes (1), keiki.payments (11), keiki.people (3), keiki.provider_accounts (2), keiki.sandbox_providers (1), keiki.sandbox_scripts (8), keiki.secrets (3), keiki.shopify (1), keiki.signals (1), keiki.skills (8), keiki.sso (7), keiki.subagent_tasks (1), keiki.trace_exporters (4), keiki.traces (6), keiki.wallet (1), keiki.webhook_deliveries (3), keiki.webhooks (7) Root calls: keiki.code_task, keiki.consult_oracle, keiki.build_tool, keiki.publish_tool, keiki.connect_mcp, keiki.read_result, keiki.talk_to_agent, keiki.check_job, keiki.cancel_job, keiki.deploy_agent, keiki.get_deployment, keiki.delete_deployment, keiki.list_agent_templates, keiki.list_channel_types, keiki.switch_organization, keiki.escalate, keiki.current_context
Run a JavaScript program against the Keiki API. `code` is an async arrow function; what it returns is the result. Inside the program `keiki` is the whole API: - grouped calls, e.g. `await keiki.agents.get_by_id({ id })`, and root calls, e.g. `await keiki.talk_to_agent({ agent, message })`; - `await keiki.search({ query?, group?, limit?, offset? })` → `{ total, offset, hits: [{ path, signature, description, readOnly, destructive }] }`. An exact path as the query returns that call's signature. Search before calling anything whose input you have not seen. A call returns the endpoint's data and throws on failure (catch it to handle it). Run independent calls together with Promise.all. The program has no network access besides `keiki`, makes at most 100 calls and runs for at most 120s. Long calls (talk_to_agent, code_task, consult_oracle, build_tool) can return { jobId, status: "running" }: follow with keiki.check_job({ jobId, waitMs }) here or in a later program, or cancel_job({ jobId }) to request a stop. Calls with destructive: true delete, spend money or change who has access. Show the user what the program will do and get their confirmation before running one. Results over 60,000 serialized characters are stored whole for one hour and return { resultId, sizeChars, preview, note }. Read windows with keiki.read_result({ id: resultId, offset, length }) (at most 50,000 characters) or search with { id: resultId, pattern, maxMatches }. Storage is capped at 5M characters per result and 20M live characters per user; oldest results may be evicted. Storage failure is an error, never clipped data. Use keiki.connect_mcp({ agentId, presetId, waitMs? }) after saving a preset. Open a needs_consent authorizationUrl in the user’s browser, then poll again with waitMs. Header credentials belong on the linked agent page, never in the program or chat. Example: async () => { const { hits } = await keiki.search({ query: "agent versions" }); return hits.map((hit) => hit.signature); } Groups: keiki.agent_goals (1), keiki.agent_groups (12), keiki.agent_templates (1), keiki.agentmail (2), keiki.agents (86), keiki.approvals (3), keiki.audit (2), keiki.auto (1), keiki.axiom (1), keiki.billing (18), keiki.blog (2), keiki.capability_tools (1), keiki.channel_access_requests (2), keiki.conversations (14), keiki.dashboard (1), keiki.evals (16), keiki.identities (53), keiki.insights (1), keiki.invitations (4), keiki.keys (5), keiki.lines (10), keiki.live (1), keiki.loops (1), keiki.marketplace (5), keiki.mcp (3), keiki.mcp_clients (2), keiki.members (4), keiki.models (1), keiki.notifications (2), keiki.org_mcp_integrations (5), keiki.orgs (6), keiki.outcomes (1), keiki.payments (11), keiki.people (3), keiki.provider_accounts (2), keiki.sandbox_providers (1), keiki.sandbox_scripts (8), keiki.secrets (3), keiki.shopify (1), keiki.signals (1), keiki.skills (8), keiki.sso (7), keiki.subagent_tasks (1), keiki.trace_exporters (4), keiki.traces (6), keiki.wallet (1), keiki.webhook_deliveries (3), keiki.webhooks (7) Root calls: keiki.code_task, keiki.consult_oracle, keiki.build_tool, keiki.publish_tool, keiki.connect_mcp, keiki.read_result, keiki.talk_to_agent, keiki.check_job, keiki.cancel_job, keiki.deploy_agent, keiki.get_deployment, keiki.delete_deployment, keiki.list_agent_templates, keiki.list_channel_types, keiki.switch_organization, keiki.escalate, keiki.current_context