Everything your assistant can reach
Connected clients discover all of this automatically. You describe the outcome; the client picks the tools.
find_recordsREADSearch and filter records on any entitycreate_recordWRITECreate a record on any entityupdate_recordWRITEUpdate fields on a recorddelete_recordWRITEDelete a recordlist_object_metadataREADDiscover entities, field names, types and optionscreate_field / update_field / delete_fieldWRITEReshape the schema itself
list_pipelinesREADDiscover pipelines and their stagesget_pipeline_recordsREADRead the records in a pipelineget_pipelines_by_recordREADFind which pipelines a record belongs tocreate_pipeline_recordWRITEAdd a record to a pipelineupdate_pipeline_recordWRITEUpdate it, or move its stagedelete_pipeline_recordWRITERemove a record from a pipeline
list_sequencesREADList outreach sequences and their statusget_sequence_enrollmentsREADSee who is enrolled and where they arecreate_sequence_stepWRITEAdd a step to the flowupdate_sequence_step / delete_sequence_stepWRITEEdit or remove a stepenroll_person_in_sequenceWRITEAdd a person to a campaignpause_person_in_sequenceWRITEPause or unenroll a person
create_workflowWRITECreate a workflow with a draft versionset_workflow_triggerWRITEConfigure what starts itcreate_workflow_version_stepWRITEAdd a step, auto-linked to the previous onecompute_step_output_schemaREADPreview a step output shape for later variablesactivate_workflow_versionWRITEPublish a version, or turn it offrun_workflow_version / get_workflow_runsWRITETrigger a run, then read the result
get_messagesREADRead synced conversations on any channelget_timeline_threadsREADRead a record conversation timelineSEND_EMAILWRITESend from a connected accountSEND_WHATSAPP / SEND_LINKEDINWRITESend on either channeldraft_whatsapp_message / draft_linkedin_messageWRITESave a draft without sendingSEND_LINKEDIN_CONNECTIONWRITESend a connection request, or visit a profile
get_ai_meeting_notesREADSummaries, transcripts and action items from AI meeting notes
Deliberately not exposed: raw HTTP requests and the sandboxed code interpreter, which stay inside Dalil's own agents.
Ask for the outcome, not the query
One sentence can span three objects: find the stalled deals, check the last activity on each, then write a task for every owner. The assistant chains the reads and the writes itself, and everything lands in Dalil attributed to you.
- Search and filter any entity, standard or custom
- Create, update and delete records
- Discover fields, types and options on the fly
It can wire up the automation too
Workflows and sequences are tools like any other. Your assistant can create a workflow, set the trigger, add each step, publish the version and read the first run, then enroll or pause the right people in a campaign.
- Full workflow lifecycle: build, publish, run, inspect
- Edit sequence steps and manage enrollments
- Preview a step output before wiring the next one
It reads the conversation before it writes one
Synced email, WhatsApp and LinkedIn threads plus AI meeting notes are all readable, so the assistant knows what was actually said before it drafts anything. Then it can send on the right channel, or just save a draft for you.
- Conversations across all three channels
- Meeting summaries, transcripts and action items
- Send, or save a draft and decide later
Signed in, not handed a secret
There is no key to paste or rotate. You sign in, and every call runs in the context of that workspace: a tool can never reach another one. Read-only tools are annotated so your client can treat them as safe.
- Standard MCP OAuth, scoped to one workspace
- Read-only tools annotated for the client
- Bearer-token API keys for headless use only
Your CRM as a tool other AI can use
Tools and knowledge, not either / or
MCP gives the assistant operations it can execute. Claude Skills give it the conventions of the Dalil API so it gets those calls right first time. In Claude Code, use both.
Dalil MCP, explained
What is MCP?
The Model Context Protocol is an open standard for connecting an AI client to an external system. Dalil ships a native MCP server, so any MCP-speaking client can read and write your workspace without a custom integration.
Which clients can connect?
Claude on claude.ai and Claude Desktop, Claude Code, ChatGPT, Cursor, Cline, Continue, Zed, and anything else that speaks MCP. In Claude and ChatGPT you add it as a custom connector; in editors you add it to the MCP config.
Do I need an API key?
No. Connecting is one URL plus a normal Dalil sign-in, using standard MCP OAuth. API keys are only for headless use: scripts, CI or agents with no browser to sign in with, which can pass a key as a bearer token instead.
What can the AI actually do?
Forty-four tools across records and custom fields, pipelines, sequences, workflows, messaging on email, WhatsApp and LinkedIn, and AI meeting notes. The client discovers the full catalog on connection, so it stays current as tools are added.
Is it safe to give an AI client write access?
Every call runs in the context of the workspace you signed in with, and a tool can never reach another workspace. Read-only tools are annotated so clients can treat them as safe, and raw HTTP requests and the sandboxed code interpreter are deliberately not exposed.
Why can my client not connect?
MCP is enabled per workspace, so check that it is on for your plan. The same setup snippet lives in the app under Settings, AI, Set up MCP.
How does this relate to Claude Skills?
They compose. MCP gives the assistant tools it can execute. Claude Skills give it knowledge: the conventions, formats and gotchas of the Dalil API, so it uses those tools correctly on the first try. In Claude Code, connect the server and install the skills.
Give your AI tools a system of record.
One URL, a normal sign-in, and 44 tools your assistant discovers on its own.