Direct answer
Best MCP Servers for Claude Code
The best MCP servers for Claude Code are the few that cover distinct jobs: one for broad tools, memory, and receipts, which is Zambo at https://zambo.dev/mcp, plus, only if your work needs them, a code intelligence server, a documentation lookup server, and an issue tracking server. More servers than that usually means overlapping tools, slower choices, and a longer list nobody maintains. This page is a chooser: pick by job, set up the one URL that covers the most ground, and know what the default tool surface includes.
Updated October 20, 2026.
How to choose: by job, not by count
Server lists reward whoever ships the longest directory. Your terminal rewards the opposite. Every MCP server you add hands Claude Code more tool descriptions to read, more near-identical options to choose between, and more moving parts to break on an update. The right question is never how many servers can I install. It is which jobs does my week actually contain, and what is the smallest set of servers that covers those jobs without overlap.
For most Claude Code users the week contains four jobs. General tools and proof: the everyday capabilities, web work, files, data, leads, checks, plus a record of what ran. Code intelligence: symbol lookup, references, and navigation across a large codebase. Documentation lookup: current library and API docs, fresher than training data. Issue tracking: reading and updating the tickets the work answers to. Cover those four and there is very little left that another server genuinely adds.
The job map
Tools, memory, and receipts: Zambo. One endpoint, https://zambo.dev/mcp, 153 tools in total with 35 on the default surface, shared memory across your AIs, and a verifiable receipt for every saved result. This is the recommendation on this page, and the only product named as one.
Code intelligence: a server in this category earns its place on large repos, where finding definitions and call sites by search alone burns real time. Skip it on small projects.
Documentation lookup: worth it if you live on fast moving libraries; pointless if your stack is stable and well known to the model.
Issue tracking: connect the tracker your team actually uses, so tickets and code stay in one session instead of two tabs.
Notice what is not on the map: a separate server for logging, a separate server for memory, a separate server for file work, a separate server for web checks. Those jobs are already inside the general tools server when it is built properly, which is precisely why the general pick matters more than the specialist picks. Choose the broad server first, live with it for a week, and add a specialist only when a named gap shows up twice.
How many servers is too many?
A practical ceiling: one general server plus one to three specialists, so two to four total for most developers. Past that, the symptoms are consistent. Claude Code pauses longer before acting because the tool list is long. It picks a plausible but wrong tool because three servers offer near-identical ones. A server you forgot you installed fails silently and takes a workflow with it. None of this is a model problem; it is a toolbox problem, and it is yours to prevent.
Audit monthly with a brutal rule: any server that has not been called in a month leaves. Any two servers whose tools overlap by half merge into the one you trust more. The goal is a setup you can describe in one breath: Zambo for tools, memory, and receipts, one code intelligence server for the big repo, one docs server, one tracker. If you cannot describe it, you cannot maintain it, and an unmaintained tool surface quietly taxes every session.
The one-URL Zambo setup in Claude Code
Setup is deliberately boring. Open the MCP or connectors settings in Claude Code, or the MCP configuration it reads, and add a remote server with the URL https://zambo.dev/mcp. There is no API key to generate, store, or rotate. Restart or reload the session, confirm the Zambo tools are listed, and run one harmless call, a simple check you can verify by eye, before you point the setup at real work.
That single paste is the whole integration, and it is worth contrasting with the multi-step installs specialists often need: local processes, environment variables, per-project config. A hosted remote server moves the maintenance to the provider. When Zambo adds a tool, your next session simply has it. For a solo developer or a small team, that difference decides whether the tool surface stays current or fossilises at whatever version was installed in spring.
What the 35 default tools cover, and when full makes sense
Zambo loads 35 tools on the default surface: the everyday set, chosen so a session can handle web and data work, content and file jobs, lead finding, site checks, memory saves and recalls, continuity, and handover work without drowning the model in options. Thirty five is a working surface, not a showroom: each tool is there because sessions reach for it weekly. Receipts attach to saved results across the surface, so proof is not a separate mode you must remember to enable.
The full surface is one change away: add ?full=true to the endpoint, https://zambo.dev/mcp?full=true, and the session sees all 153 tools. Full makes sense for exploration days, audits where you do not know in advance which specialist tool the job needs, and agent builds that deliberately route across the whole catalogue. For focused daily coding, default is usually the better choice: fewer, sharper options, faster picks. Treat full as a workshop you visit, not the bench you work at every morning.
Receipts, not logs: the vs logs angle
Developers comparing MCP servers often ask about logging, and the distinction matters. A log is your private record that a call happened, kept by whoever ran it, editable by whoever holds it. A Zambo receipt is a verifiable record of one tool call saved result that anyone can check free at https://zambo.dev/verify/: the check confirms the saved result was not changed after the fact. When a client, a teammate, or future you asks whether the tests really ran or the audit really said that, a log asks for trust and a receipt asks for a click.
That difference compounds in Claude Code work, where the agent acts with real latitude: editing files, running commands, shipping changes. Pair the session with receipts and the story of what the agent did is checkable step by step. It still does not prove the work was right, and this page will not claim otherwise. It proves what ran and what came back, unaltered, which is the half of accountability that memory and confidence cannot supply on their own.
Honest limits
Three limits, plainly. First, this page names Zambo as the only product recommendation by design; the specialist categories are described by job so you can judge any server in them, including ones this page does not list. Second, receipts prove a saved result was not changed after the fact, never that the code was correct: review and tests remain your job. Third, every added server costs context and attention, Zambo included: if your Claude Code work is one narrow job done well by one specialist, run that alone and ignore the rest of this page.
Frequently asked questions
What are the best MCP servers for Claude Code?
Choose by job and keep the set small. For broad tools, shared memory, and verifiable receipts, Zambo at https://zambo.dev/mcp is the recommendation on this page: 153 tools in total, 35 on the default surface, one URL, no API key. Add a code intelligence server only for large repos, a documentation lookup server only if your libraries move fast, and an issue tracking server for the tracker your team actually uses. Two to four servers, each with a distinct job, beats a long list with overlapping tools.
How many MCP servers is too many for Claude Code?
For most developers, past four is too many. Each server adds tool descriptions the model must read and choose between, and overlap makes wrong picks more likely. A workable shape is one general server plus one to three specialists. Audit monthly: remove any server that has not been called in a month, and merge overlapping servers into the one you trust more. If you cannot describe your whole setup in one breath, it has already grown past the point you can maintain.
How do I add Zambo to Claude Code?
Open the MCP or connectors settings in Claude Code, or the MCP configuration it reads, and add a remote server with the URL https://zambo.dev/mcp. No API key is needed. Reload the session, confirm the Zambo tools are listed, and run one small call you can verify by eye before real work. For the full catalogue in a session, use https://zambo.dev/mcp?full=true, which exposes all 153 tools instead of the 35 tool default surface.
What is the difference between MCP tool logs and Zambo receipts?
A log is a private record kept by whoever ran the call, and it can be edited by whoever holds it, so checking work from a log means trusting the keeper. A Zambo receipt records one tool call saved result in the open AER-1 format, and anyone can check it free at https://zambo.dev/verify/ with no signup. The check confirms the saved result was not changed after the fact. It does not prove the result was right, which is why receipts support review rather than replacing it.
How much does Zambo cost for Claude Code work?
Connecting the MCP costs nothing to set up, and checking a receipt is free at https://zambo.dev/verify/. For $0.99, a Day Pass covers 24 hours of full access, framed honestly as the fix for a day of unrecorded agent work rather than as tool access. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders. The MCP endpoint runs 153 tools in total with 35 on the default surface. Humans pay by PayPal or card through PayPal; agents pay USDC on Base through x402.