Skip to main content
Glama
agentreadycrm

agentreadycrm

Official

Agent Ready CRM

A CRM for AI agents, served as an MCP server: organizations, contacts, interactions, deals, and the evidence behind them. An agent resolves a company or a person before it writes, logs every touch, moves deals through a pipeline, and records the facts it found with the quote and the URL that prove them. A person sees the same records in views that render inside MCP Apps hosts.

It runs on your own machine: one SQLite file, no account, no cloud service. A hosted service at mcp.agentreadycrm.com, serving the same tools to a team, is coming soon.

Install

Run it with uv (Python 3.13 or later):

uvx agentreadycrm

That serves MCP over stdio. To add it to Claude Code:

claude mcp add agentreadycrm -- uvx agentreadycrm

For Claude Desktop and other hosts that install MCP bundles, the bundle @nimblebraininc/agentreadycrm is available from mpak. It runs the same package and asks where to keep the database.

Any other MCP client that starts a stdio server can run uvx agentreadycrm the same way. To serve HTTP on 127.0.0.1 instead, run uvx agentreadycrm --http; HTTP requires a bearer token, which uvx agentreadycrm --print-token prints. See docs/local.md.

Related MCP server: agent-crm

Where your data lives

Everything is in one SQLite file, crm.db, in your user data directory:

OS

Default location

macOS

~/Library/Application Support/agentreadycrm/crm.db

Linux

$XDG_DATA_HOME/agentreadycrm/crm.db (~/.local/share/... when unset)

Windows

%LOCALAPPDATA%\agentreadycrm\crm.db

Set PEOPLE_LOCAL_DB to use another file. Keep it on a local disk, not a network share. Several clients may use the same file at once. Upgrades, backups and the HTTP token are covered in docs/local.md.

Local or hosted

Local (this package)

Hosted (coming soon)

Runs

On your machine, started by your MCP client

At mcp.agentreadycrm.com

Data

One SQLite file you own

A managed Postgres database

Who

One person

A team, each signed in

Tools, views, skills

The same

The same

Both serve the same core: a client sees the same tools, schemas, views and skills, and only where identity comes from and where the data lives differ.

Tools

46 tools, in these families:

Family

Tools

Resolve and import

find_or_create_organization, find_or_create_contact, import_organizations, import_contacts (up to 500 rows a call)

Read and list

get_contact, get_organization, get_opportunity, list_contacts, list_organizations, list_opportunities, list_interactions, count_records, resolve_external_refs

Change

update_contact, update_organization, update_opportunity, log_interaction

Duplicates, hide and delete

merge_contacts, merge_organizations, archive_contact, archive_organization, delete_contact, delete_organization

Deals and the pipeline

create_opportunity, move_opportunity, get_pipeline, define_pipeline_stage, update_pipeline_stage

Custom fields

define_field, update_field, list_fields

Lookalike companies

link_similar_organizations, list_similar_organizations, promote_organization

Evidence: signals

record_signal, withdraw_signal, update_signal, get_signal, list_signals, list_signal_types, define_signal_type, update_signal_type

Evidence: proof points

record_proof_point, verify_proof_point, get_proof_point, list_proof_points

A few rules shape all of them:

  • Resolve before you write. find_or_create_* matches by external reference, then domain or email, then name, and a match fills only empty fields, so repeating a call converges on one record.

  • Every result carries a next_step telling the agent what to do with it, and every failure is one error envelope with a next_step that names the fix.

  • Listings query, they don't walk. list_* take a free-text query, a filter and sort on any core, custom, related or rollup field, and columns, and page with a cursor.

  • Nothing is removed. A delete marks a record so no read finds it; archiving is the reversible hide.

The server also serves two skills (skill://people/SKILL.md and skill://people-evidence/SKILL.md) with worked examples for hosts that load skills.

Views

Four views, ui://people/contacts, organizations, opportunities and settings, render in hosts that support MCP Apps. Each record has an address (people://contacts/<id> and so on) that a host can open in a view. The views read and change data only through the same tools.

More

  • docs/architecture.md: runtimes, the data model, deletion and archiving, the surface, errors and annotations.

  • docs/local.md: the database file, several processes, the HTTP token, upgrades, backups, and what differs on SQLite.

  • CONTRIBUTING.md: building and testing.

Licence

MIT. See LICENSE.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to interact with a CRM covering companies, people, leads, deals, and more, with role checks, scoped agent keys, approval gates, and a shared audit trail.
      AGPL 3.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to manage CRM data including companies, contacts, prospects, pipelines, forecasts, and tasks via typed MCP tools, with local SQLite storage and a JSON CLI.
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI assistants to manage an outreach CRM via SQLite, including importing contacts from Apollo/Crunchbase, fuzzy matching, seniority ranking, drafting and scoring emails, and tracking sends/replies through pluggable email providers.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to read and maintain a personal CRM as an Obsidian vault of plain Markdown notes, covering people, organizations, emails, meetings, tasks and topics with duplicate detection and source-backed facts. Runs entirely locally, with a SQLite index derived from the notes so lookups stay fast while the vault remains the source of truth.
      MIT