Key takeaways:
- Outreach’s MCP server exposes around 44 tools spanning conversation intelligence, sequences, prospects, accounts, opportunities and configuration, which is the deepest catalog of any sales engagement MCP server we have reviewed.
- Access requires the Amplify add-on with active credits, plus an active licensed Outreach seat, so it is not available to every Outreach customer by default.
- An admin must switch the server on per organization, and the create and delete permissions are separate toggles, with delete off by default.
- Outreach acts as both an MCP server and an MCP client, which is a genuine architectural difference from its competitors.
The Outreach.io MCP server is the most enterprise-shaped in the sales engagement category, and both halves of that sentence matter.
The tool catalog is deep enough to run real revenue workflows, reaching into MEDDPICC fields on opportunities and searchable call transcripts. The access model is deep enough to need an admin, a licensed seat and a paid add-on before anyone connects.
This Outreach.io MCP review covers what the server exposes, how an admin turns it on, and the limits worth understanding before you plan around it. We compete with Outreach directly and run our own MCP server, so treat the comparison section accordingly and take the factual sections on their own terms.
What Outreach MCP Is
Outreach MCP Server implements Anthropic’s Model Context Protocol, the interoperability standard that lets one vendor’s data reach another vendor’s assistant. In Outreach’s framing, the platform acts as a producer of knowledge and a facilitator of actions for MCP clients.
The unusual part is that Outreach plays both roles. Alongside the server, it ships MCP Connectors that let Outreach act as a client, pulling knowledge and actions from other systems into Outreach itself. Most vendors in this category have built one direction only.
That bidirectionality is a real architectural choice rather than a marketing line. It means Outreach can be the place where context from other tools lands, as well as a source other assistants query, which suits organizations trying to make Outreach the system of record for revenue activity.
The design also reflects where the category is heading. The interesting question in AI-driven outbound has moved from whether a platform has AI features to whether its data is reachable by whatever assistant a team already uses, and a server plus a client is the most complete answer a vendor can give to that.
Who Can Actually Use It
The prerequisites are stricter than most, and they are the first thing to check.
Your organization needs the Amplify add-on enabled with active credits. Outreach’s own documentation is direct about the consequence: if you do not see the setup options described below, you may not have Amplify, and the next step is asking your account executive about it.
Beyond that, the user must be active with a licensed seat on at least one Outreach instance, and the client itself may require its own licenses. The server is available to all Outreach customers to connect to, but Amplify is what makes it usable.
What Outreach MCP Can Do
The catalog runs to roughly 44 tools across eight groups. What distinguishes it is not the CRM-style coverage, which several servers now have, but the conversation intelligence and the deal fields.
| Group | Representative tools | What it covers |
| Conversational intelligence | kaia_meeting_search, kaia_meeting_fetch, emails_search |
Search calls by keyword, topic, participant or transcript text, then fetch summaries and transcripts |
| Workflow | sequence_search, sequence_create, sequence_add_prospects, calendar_events_fetch |
Find, build and populate sequences, and retrieve scheduled meetings |
| Task management | task_create, task_search, task_priority_fetch, task_theme_search |
Create and find tasks across call, email, LinkedIn, SMS and in-person types |
| Accounts | account_search, account_answer_question, account_create |
Look up accounts and ask analytical questions across records and transcripts |
| Prospecting | prospect_search, prospect_get_by_id, prospect_create |
Find and create prospects, including lookup by CRM ID |
| Deals | opportunity_search, opportunity_answer_question, opportunity_stage_fetch |
Query deals, including MEDDPICC methodology fields on request |
| Users and teams | current_user, current_org, job_role_fetch, team_search |
Resolve identity, org defaults and role structures |
| Configuration | filter_schema_fetch, input_fields_fetch, stage_fetch, ruleset_search |
Discover valid fields, filters and operators before constructing a query |
Two design decisions stand out. The first is account_answer_question and its opportunity equivalent, which analyze the record plus meeting transcripts and related data to answer a general question rather than returning rows. That is a higher-level abstraction than most servers offer, and it is why Outreach’s sample prompts read like “tell me Contoso’s concerns around our solution and how I should combat that.”
The second is the configuration group. Tools such as filter_schema_fetch are required before constructing filters on search tools, returning standard fields, custom fields with their validation types, and the valid operators for each. It adds a round trip, and it is what makes filtering reliable across organizations with different custom-field setups.
Outreach has not exposed any resources or prompts, so the server is tools-only. It also notes that response quality depends on the richness of the data already in Outreach, which is a fair caveat and applies to every server of this kind.
Sequence creation is more constrained than the tool name suggests. sequence_create sets name, description, sales motion, owner, sequence type, share type, schedule, throttle and ruleset, with the owner defaulting to the current user and the throttle defaulting to 50 adds per day. What it does not do is author the message content, so the branching logic behind conditional sequences still gets built in the interface rather than through conversation.
Bulk operations behave asynchronously, which is the right design and changes how you prompt. sequence_add_prospects creates a batch operation and returns a batch ID to poll, so an assistant that reports success has confirmed the batch was accepted rather than that every prospect landed.
How to Set Up Outreach MCP
Setup is an admin task first and a user task second. Nobody can connect until the organization-level toggle is on.
Enabling the Server as an Admin
The server is off until an administrator switches it on, and the setting is per organization, exposing only that instance’s data.
- Log in to Outreach as an admin.
- Click Administration (settings), then Organization, then Org Info.
- Navigate to MCP Server and toggle it on.
- Review the two optional permission settings, which control whether create actions are allowed (on by default) and whether delete actions are allowed (off by default).
That third and fourth step is the part worth pausing on. Outreach ships create enabled and delete disabled, which is a defensible default, and it means an assistant can add prospects and sequences from day one unless you decide otherwise.
Connecting a Client
Once the server is on, users authenticate with their own Outreach credentials plus any additional factors the organization requires. Any AI application supporting the protocol, user-level auth and Dynamic Client Registration can connect, and Outreach maintains an up-to-date listing in its marketplace under MCP Connectors.
Supported clients include Claude Desktop and Claude on the web, ChatGPT and Codex, Salesforce Agentforce through Apex invocable actions, Microsoft Copilot via Copilot Studio and Microsoft 365 apps, Gemini through the CLI, AWS Bedrock and Amazon Quick, and IDEs such as Visual Studio Code.
One practical gotcha causes most failed connections. Most clients launch your default browser to authenticate and pick up whichever account is already signed in there. If the wrong instance connects, sign out in that browser, sign back in with the correct account, and retry.
A Note on Agent Identity
Outreach does not support client-auth-only clients. It requires an active user credential, SSO or otherwise, and applies the profile assigned to that user account.
If you plan to run an autonomous agent against Outreach, that agent identity must be an active, licensed user in Outreach. This is a licensing consideration as much as a technical one, and it is easy to miss when scoping an agentic project.
The approach is aligned with where the standards have landed. Outreach follows the MCP authorization standards, which build on OAuth 2.0 Protected Resource Metadata, defined in IETF RFC 9728 as the metadata format a client uses to discover how to interact with a protected resource. Requiring a real user identity rather than a shared service credential is the stricter and safer reading of that model.
How Teams Use Outreach MCP
The prompts Outreach highlights cluster into three jobs, and they are a fair guide to where the server earns its keep.
Call preparation is the strongest. Asking the assistant to prepare you for the next call with an account pulls the record, recent meetings and transcripts together into a briefing, which is the sort of task that otherwise takes fifteen minutes of clicking.
Objection handling is the second, and it depends on the question-answering tools rather than on search. Asking what an account’s concerns are, and how to address them, works because account_answer_question reasons across the record and the transcripts rather than returning a list of rows.
Post-meeting follow-through is the third. Generating a summary of features discussed on the last call, then creating the follow-up tasks that come out of it, chains conversation intelligence into task management in a single request.
What the server does not do is reach outside Outreach for new prospects, so top-of-funnel work still needs a data source. It also has no LinkedIn execution surface, which is a real gap for multichannel teams and one worth checking against dedicated LinkedIn outreach tools before assuming one server covers every channel.
We can attach a LinkedIn account to a sequence from our own tools, and we should say what that costs rather than implying it is included. LinkedIn automation bills separately from the base plan, at $69 a month for every account you connect.
Outreach MCP Pricing
Outreach does not publish list pricing, and the MCP server inherits that. What we can state precisely is the gating rather than the cost.
Access requires the Amplify add-on with active credits, which is a separate purchase from a standard Outreach subscription. Outreach’s documentation routes you to your account executive to find out whether you have it and what it costs, which means budgeting for MCP access requires a conversation rather than a pricing page.
That opacity is worth weighing against alternatives when you are comparing servers. Several competing platforms include MCP access in their base subscription or in a free trial, and the total cost of an Outreach MCP rollout is not something you can determine without a quote.
The add-on structure also means the MCP decision is rarely made on its own. Amplify carries capabilities beyond the server, so teams generally evaluate the whole package rather than the protocol access in isolation, and our Outreach review covers the platform surfaces that come with it.
